Stanford AA274A Principles of Robotic Autonomy | Autumn 2019 | The Robot Operating System (ROS)

斯坦福 AA274A 机器人自主性原理 | 2019年秋季 | 机器人操作系统（ROS） Benoit Landry 博士介绍了机器人操作系统（ROS）的架构与运行范式，阐释了机器人软件如何通过模块化以及发布者-订阅者设计模式来应对复杂性。他概述了 ROS 核心原语的机制，包括 ROS Master、节点、话题、自定义消息以及 Catkin 工作空间规范，并将异步发布-订阅架构与过程式脚本进行了对比。通过命令行诊断和实时的多节点计算机视觉流水线，Landry 展示了硬件驱动程序与算法如何在分布式网络中实现点对点通信。最后，本讲探讨了基础机器人工具，包括 launch 文件的编写、URDF 机器人建模以及 Gazebo 中的物理仿真，强调了仿真环境与物理硬件之间的软件一致性。 Benoit Landry 博士介绍了作为斯坦福大学 AA274A 课程基础中间件的机器人操作系统（ROS），并强调其发布者-订阅者架构在现代机器人研究与工业界中不可或缺。 (0:05) ROS Master 严格充当用于在节点之间建立点对点 TCP 连接的集中式匹配注册表，其本身并不路由传感器或消息的有效载荷。 (23:21) ROS 通过 Catkin 构建系统和 URDF 运动学建模实现了多语言集成与软件可移植性，使相同的代码栈能够在 Gazebo 仿真环境和像 TurtleBot 这样的物理平台上无缝运行。 (69:42) 尽管 ROS 2 已经出现，AA274A 仍然选择了 ROS 1（Kinetic），原因在于其拥有十余年积累的成熟社区功能包、稳定的文档以及经过验证的教学可靠性。 (72:20)

斯坦福大学 Marco Pavone 教授与主讲人 Benoit Landry 博士在 AA274A 课程中阐述，机器人操作系统（ROS）本质上是基于发布-订阅（Pub-Sub）设计模式构建的网络化中间件，凭借高度模块化和成熟的开源生态，它已成为学术界与工业界管理复杂自主系统的通用基础框架。 架构模式：解耦复杂的发布订阅机制 (3:49) 与机器学习领域常见的自包含单脚本任务不同，现实中的机器人开发面临感知、规划和控制之间极高维度的软硬件协同挑战。为规避庞大单体系统的脆弱性，ROS 确立了以发布-订阅（Pub-Sub）为核心的模块化设计模式。系统将各个独立功能划分为被称为“节点”（Nodes）的操作系统进程，每个节点可以是对物理传感器（如激光雷达、IMU）的底层驱动封装，也可以是纯软件算法（如前向动力学模型或路径规划器）。节点之间通过带有名称的通道即“话题”（Topics）进行异步、解耦的多对多结构化消息通信，使不同计算模块能够独立保持各自的循环频率，从而大幅降低系统耦合度与开发难度。 通信机制：Master作为协调者而非网络瓶颈 (23:21) 针对实际运行中常见的性能认知误区，Landry 澄清了核心协调组件“ROS Master”的真实职能。ROS Master 在网络中扮演注册中心角色，维护全局参数服务器并负责节点的动态发现与身份验证。当发布者与订阅者建立通信时，Master 仅负责交换两端的网络端点元数据，协助双方协商建立底层的点对点 TCP 直连通道，随后便退出实际传输流程。由于动态自治系统中各功能节点需要根据不同场景频繁启停（例如自动驾驶系统在特定路口动态加载行人检测节点），Master 的持续运行是维持动态拓扑所必需的，但其本身绝非数据吞吐的网络瓶颈。 工程闭环：从Launch编排到Catkin与调试 (48:08) 为有效管理多节点构成的庞大系统，ROS 提供了完善的代码组织与工程构建标准。在代码实现上，工业级应用通常采用类包装结构来整合订阅回调与发布循环，并利用 rospy.Rate 机制精确适配仿真或物理时钟。为了避免逐一在终端通过命令行启动几十个独立进程，系统引入 XML 格式的 `.launch` 文件，支持跨包嵌套引用、命名空间划分以及参数动态重写，实现一键部署。在代码维护层面，Catkin 构建系统将不同“能力”（Capabilities）划分为标准化源码包，使 Python 脚本与经编译的 C++ 代码无缝混编；配合 rqt_graph 动态拓扑图、rqt_console 日志过滤以及 rostopic 命令行检测，开发者能够高效排查发布延迟与积压问题。 仿真与生态：URDF统一建模与代码无缝复用 (67:43) 统一机器人描述格式（URDF）通过标准化的 XML 树状拓扑描述刚体“连杆”（Links）与旋转/滑动“关节”（Joints），构建出适用于 Atlas 或 TurtleBot 等异构平台的统一运动学和动力学链条。这种统一抽象使代码高度解耦：同一套自主控制栈既能连接物理硬件，也可无缝挂接到高保真动力学仿真器 Gazebo 中。仿真环境下与真实机器人共享完全一致的话题接口和节点交互逻辑，这不仅使学生在不接触昂贵硬件的情况下即可验证运动规划控制算法，更极大保障了算法由虚拟仿真向实体机器迁移时的零代码返工与高复用性。 Landry 指出，尽管 ROS 2 自 2007 年初代发布以来带来了面向现代机器人更深层底层的架构演进，但在教学与原型设计中，成熟稳定、文档详尽且拥有海量开源包的 ROS 1 仍是掌握分布式自主系统心智模型的最佳基石。

[69:42] Dr. Benoit Landry: If it doesn't work in simulation, it's probably not going to work on the real robot. And then if it works in simulation, it might work on the real robot. [69:42] Dr. Benoit Landry: If it works in simulation, but now to make it work on the real robot you have to rewrite everything from scratch, the simulation really gave you very little.

[0:05] So today we're going to talk about Ross, [0:08] the Robot Operating System. This is the [0:11] software environment in a way that we'll [0:14] be using in this class to architect the [0:17] autonomous stack that we will be [0:19] incrementally building throughout the [0:21] quarter. I cannot under emphasize or [0:24] over emphasize actually the importance [0:26] of Ross in the robotics community. If uh [0:30] you will eventually become a roboticist [0:32] and have a career in robotics, the [0:34] chances that you will be using Ross will [0:36] be extremely high. [0:37] And one of That's one of the main [0:39] objective of this class is for you to [0:40] familiarize with the [0:42] with Ross, the Robot Operating System. [0:45] So the lecture today is going to be [0:46] given by one of the PhD student in my [0:48] lab, Benoît. [0:50] He's a [0:51] senior PhD student in my lab. He's a [0:54] veteran in Ross, in the Robot Operating [0:56] System. So he's first going to give you [0:59] overview of what Ross is really what [1:03] Ross really is. He's going to give you a [1:06] demonstration of [1:09] of Ross. [1:10] And of course in the following weeks, by [1:12] attending the sections, you will become [1:15] increasingly more exposed to Ross to the [1:16] level that you'll be able to build your [1:19] own autonomous stack in Ross and finally [1:21] deploy it on on the robotic platform [1:23] that we will be using in this class. [1:26] So any questions before we start? [1:31] No? Good. [1:33] >> Oh, okay. Thank you, Marcus. [1:36] Um [1:37] Yeah. [1:39] >> Okay. [1:40] Huh. Nice. [1:42] Haven't done a CPD before, so [1:44] Um cool. So yeah, so today we're going [1:46] to talk about the Robot Operating [1:48] System. And that's a bit of a different [1:49] lecture than what you're going to do for [1:51] the rest of the quarter. So in case I [1:52] know you guys are still kind of, you [1:54] know, shopping around for classes and [1:55] everything. [1:56] So this this might be just slightly [1:57] different than the rest of the quarter, [1:59] but it's still very important class. The [2:00] reason is because most of the classes [2:02] that that Professor Pavone is going to [2:04] teach are about algorithms, and today [2:06] we're talking about software, not [2:07] necessarily algorithms in general. Um [2:10] and so, we'll dive in. [2:13] Uh so, writing software for robotics uh [2:15] turns out it's actually an extremely [2:16] complicated task. This is a picture I [2:18] really like. Uh it is actually from the [2:21] DARPA uh [2:22] Robotics Challenge which happened a [2:23] couple of years ago, and this is the [2:25] team that is operating a robot that is [2:27] the Atlas robot. It's actually the the [2:29] humanoid robot that you see there, uh [2:31] and there's a whole bunch of people in [2:32] the room, and they're all kind of [2:34] monitoring. You can see I think there's [2:35] what, eight screens over there, uh and I [2:37] promise you they're all needed, and [2:38] there's probably more screens that [2:39] you're not seeing uh that that are doing [2:42] different things. And that's just to [2:43] basically get that robot not to do [2:45] anything that complicated. It's [2:46] probably, you know, trying to like walk [2:48] upstairs or doing like just not hitting [2:51] things. Um so, [clears throat] it [2:53] actually it's very complicated to run [2:55] robots, and it's obviously very [2:56] complicated to write software that runs [2:58] robots. Uh and so, [3:00] >> [clears throat] [3:01] >> uh and and something to keep in mind is [3:02] that [3:03] uh and and one of the reason we're [3:04] emphasizing this is actually that [3:06] uh you will write, you know, real [3:07] robotics software in this class over the [3:09] quarter, and uh this is probably going [3:12] to be the class where you're going to [3:13] interact with software that's the most [3:15] complicated you've ever seen. Uh it [3:17] doesn't mean it's going to be [3:18] overwhelming. It we have like we're [3:19] going to take you through it, and uh [3:21] there's there's a Professor Pavone has [3:23] kind of, you know, uh uh [3:26] has an expertise now in like taking [3:27] people through kind of the process of [3:29] learning this, uh but but it's it's [3:31] probably way more complicated than [3:33] software you interacted with in, you [3:34] know, CS 229 and ML where, you know, [3:36] everything was self-contained in one [3:37] little script, and you're just running [3:38] it, and it kind of works. Uh and it [3:40] might depend on one library. Here, [3:42] you're going to have a lot of moving [3:43] parts, uh and it's going to be just a [3:45] little bit more complicated, uh but [3:46] we're going to teach you the tools of [3:47] how to deal with that. [3:49] So, in general, uh how people deal with [3:51] the complexity of robotics uh system is [3:55] with a very popular strategy which is [3:56] modularity. [3:58] So basically the idea is is pretty [4:00] simple is that you know we're going to [4:01] basically break down this very [4:03] complicated system instead of you know [4:04] writing it as one big program we're [4:06] going to build it we're going to break [4:07] it down in kind of separate little [4:08] modules. [4:10] And the way that we're going to do this [4:11] is by following something called a [4:12] design pattern. So I don't know who here [4:15] has taken I guess principle of software [4:16] engineering or or a class like that that [4:18] probably is taught at Stanford. That [4:21] define what a design pattern is. So [4:23] there's this [4:24] you know definition that I stole from [4:25] Wikipedia which is is basically just a [4:28] reusable solution. So just like a [4:30] way of designing solution that is very [4:32] very general and that can address kind [4:35] of similar problems that keep coming up. [4:37] So the problem that keeps coming up in [4:38] robotics is that these [4:41] these pieces of software are very [4:42] complicated and so we've got to break [4:43] them down in modules and then like the [4:46] way to do that is we're going to use a [4:47] specific design pattern. [4:49] The specific design pattern that we're [4:51] going to use here that we're going to [4:53] talk about today is the pub sub design [4:55] pattern. So what that means is basically [4:57] we're going to take again our software [4:59] we're going to break it down to [5:00] different components [5:01] and then what that design pattern is [5:02] tell us to do is to basically define [5:04] something called a topic. And what topic [5:07] are is is a little bit you can think of [5:09] chat rooms. I might be [5:11] too old now to teach this but [5:14] I don't know if you people still have [5:15] chat or still remember what chat rooms [5:16] were like but they're basically you know [5:18] these like forum essentially where [5:19] everyone could just you could go on a [5:21] chat topic let's say cars and everyone [5:22] could just write anything and then [5:24] anyone who's following the topic cars [5:25] could just get the post that someone [5:27] else's is listening to. So we're going [5:28] to define these topics [5:30] and then each each of the modules now [5:32] that we've broken down our software into [5:34] can do one of two things. It can either [5:36] publish to a topic so that means send a [5:39] message on a topic or it can subscribe. [5:41] So that means that it will say that [5:43] anything anyone who's publishing [5:45] anything on a specific topic I want to [5:47] get a message from there. So, that's a [5:49] design pattern that basically defines [5:50] the design pattern. And that design [5:52] pattern is something we call pub-sub. [5:53] So, just a little diagram of that. Um [5:56] the client are like those modules that [5:58] I'm talking about. The topics are just [6:00] kind of like this abstract idea, which [6:02] are, you know, Slack channels would be a [6:04] topic. Uh that might be the more modern [6:06] equivalent that you guys would know. Uh [6:07] and then the you know, the messages [6:09] being passed passed around. Um there are [6:12] many different ways of implementing this [6:14] design pattern. Uh that's something to [6:16] keep in mind. And you also already use [6:19] uh [6:19] pub-sub a lot uh every day. And you have [6:22] a question? [6:26] >> In practice, can multiple clients [6:27] publish to the same topic? [6:29] >> Yeah. Yeah, so the the in this specific [6:32] example, you only have one client that [6:33] publishes and two of them that are [6:35] subscribing. But really, pub-sub allows [6:37] like anyone to publish and anyone to [6:38] subscribe. Um [6:40] and so, there's already actually a very [6:43] common example of pub-sub that you're [6:44] already using every day. I don't know if [6:46] someone can come up with an example. [6:49] Is there something that I'm thinking [6:50] about? [6:52] Reddit would be Yeah, on a thread. That [6:53] would probably be a good pub-sub. If you [6:55] subscribe to a thread, uh [6:57] yeah, the back. [6:57] >> Are you giving this lecture? [7:00] >> Interesting. [7:02] Yeah, maybe. That would be That's an [7:04] interesting example. Yeah, if I if maybe [7:06] if I let everyone talk, uh then then it [7:09] would be something like pub-sub. The [7:10] thing I had in mind uh is is actually [7:12] mailing lists. Uh you guys are probably [7:14] signed up to like your volleyball club [7:16] mailing list for or the Stanford like [7:17] cycling mailing list or something. [7:19] That's pub-sub. So, you have this topic [7:21] that is a cycling mailing list, right? [7:23] And then anyone can send an email to [7:25] this and everyone else who subscribed to [7:27] the like cycling mailing list will get a [7:29] message that about like a bike that you [7:31] want to sell or something like that. So, [7:33] pub-sub is everywhere, uh not just in [7:34] robotics, not just in ROS. [7:37] Uh [7:38] And there are alternative ways of [7:39] obviously dealing with that modularity [7:41] and that complexity. Uh there's [7:43] different things. So, there's for [7:44] example, you could do request-reply so [7:47] that you could [7:48] design or data binding so that you could [7:51] have like a shared [7:52] uh [7:53] a shared uh [7:55] basically object that, you know, you're [7:56] sharing data over. So, you could think [7:58] an example if we're taking the mailing [7:59] list example and we're implementing it [8:01] as with a data binding pattern, that [8:02] would be we would have for example like [8:03] a a chalkboard in a hallway and we're [8:05] all writing things on it and then [8:07] everyone is kind of seeing what everyone [8:08] else is writing on it. So, that would be [8:09] a different way of doing doing things [8:11] than doing the pub/sub. Uh so, that's [8:13] just something to keep in mind. That's [8:14] like a choice that we're deciding to [8:16] basically teach you is like pub/sub is [8:18] is one one way of dealing with that [8:20] complexity. [8:22] Um and I [8:23] let's just start on what is ROS, which [8:25] is really what this lecture is about. Uh [8:26] so, that sounds like a little maybe a [8:28] little contrived way of of going about [8:30] it. Um but, the thing to really [8:33] understand and that's like really what I [8:35] think if there's one thing you kind of [8:36] have to remember from today, you're [8:37] probably not going to remember the [8:38] syntax that I'm going to show you, but [8:40] just remember that what the pub/sub [8:42] architecture is and then the fact that [8:45] ROS, the Robot Operating System, is just [8:47] an implement at at its core, at its most [8:49] most basic definition, is an [8:51] implementation of the pub/sub design [8:53] pattern. [8:54] So, it's really just a [8:56] a very like someone basically took that [8:57] design pattern and implemented it. [8:59] Uh it's it's specifically geared to to [9:01] robotics and it's network aware, but all [9:03] of these are kind of like uh a little [9:05] bit, you know, fuzzy things. [9:08] Uh and then there's another thing to so, [9:10] the the the part where it gets a little [9:11] complicated is that, you know, if you're [9:13] trying to be like very pedantic, that's [9:15] like you'll say that it's just an [9:16] implementation of pub/sub. Um but, [9:18] usually in practice and as you know, [9:19] this is a very practical class, so we're [9:21] going to try to teach you a little bit [9:22] also kind of what people mean when they [9:24] say ROS. Uh they don't usually just mean [9:26] the pub/sub implementation. They also [9:28] usually mean the set of tools that [9:30] people have developed over time that use [9:33] ROS. [9:34] Uh and by ROS like by that I mean that [9:35] use like this pub/sub [9:37] implementation. So when when people are [9:39] saying well I'm using ROS, that may mean [9:41] that they're using this pub sub [9:42] implementation, but that probably also [9:44] means that they're using some of the you [9:45] know super popular packages that that [9:47] are supported in ROS. So that's like you [9:50] know the very kind of like everyday [9:53] definition is this set of tools, but if [9:55] you really want to be very strict about [9:57] your definition is just an [9:58] implementation of pub sub. [10:02] Um and there are alternatives to ROS. [10:05] Although I mean some of them will [10:06] qualify as an alternative only for part [10:07] of it. For example LCM is a more [10:09] lightweight version. Drake is also an [10:11] alternative more for like the tools [10:12] part. [10:14] And there's there's plenty of other [10:15] other things to do. So ROS is not the [10:17] only solution to to or the only [10:20] implementation of this pub sub pattern [10:21] or the or of these these alternative [10:24] design pattern. But why why then is ROS [10:26] so popular? [10:28] For a couple of reasons. I mean it's not [10:30] it's not that it's like faster or that [10:32] it uses some magic you know parallel [10:33] processing or anything like that. [10:35] There's very kind of like down to earth [10:36] reason why ROS is so popular in the [10:37] industry. [10:39] One of the one of them is just because [10:41] people generally don't want to reinvent [10:43] the wheel. So if someone has already [10:46] done kind of a pretty good visualizer, [10:48] it's probably a good idea to just use [10:50] that instead of spending you know your [10:52] like half of your seed fund [10:54] to just develop that. So there's [10:56] definitely just this kind of reusability [10:58] part that people definitely bank on. ROS [11:00] has been around for a while. That's one [11:02] of the first one. And so maturity of the [11:04] software is just you know it's a common [11:05] thing that people will look for. [11:08] And also it's just adoption essentially. [11:11] There's a lot of like very good group of [11:13] roboticists that use ROS. And so and [11:16] they publish a lot of open source [11:17] software. So software is just free for [11:19] you to grab online. [11:21] And so because of that it's it's it's [11:23] nice to be able to use to basically you [11:26] know leverage other people's work. And [11:28] if they happen to have implemented [11:29] everything on top of ROS then you can [11:31] use that very easily. [11:32] Uh [11:33] but obviously we're not the here. Uh [11:35] this is a class, so why are we using it? [11:37] Um for a couple reasons. I mean, [11:40] because it's used in industry, we'll use [11:41] it here uh just because it prepares you [11:43] pretty well. Uh I think to to kind of [11:45] the reality that that is going to be [11:46] there. I [11:48] uh [11:49] can at least vouch for the fact that if [11:52] you're going to work on a real robot, [11:54] you will maybe not use ROS, but you will [11:57] almost almost guarantee that you will [11:59] work on something that you will use [12:00] something that is very similar to ROS. [12:03] Uh and that's going to be like a very [12:04] good thing for you to just kind of [12:05] always fall back on on in terms of like [12:07] what to expect, uh how things are [12:08] generally designed. Uh [12:11] and it's also just an interest policy [12:13] for you. Uh we're just, you know, [12:14] teaching you kind of the most popular [12:16] thing, uh the most stable software, [12:19] uh and and yeah, what's closest to [12:21] basically an industry standard. It also [12:22] means that there's a ton of resources [12:24] online. So, we can only teach so much [12:26] and and the TAs are are doing a great [12:28] job in like bringing up material and [12:30] putting material on the website, uh but [12:31] that's going to be limited and you're [12:33] going to have questions that are going [12:33] to be outside of the scope of [12:34] necessarily, you know, the notes that [12:36] that Professor Pavone writes up. Uh but [12:39] there's, you know, literally books uh [12:41] that are written from people from [12:42] Stanford and from outside of Stanford on [12:44] on this topic. So, there's a ton of [12:46] documentation. That's not necessarily [12:47] true for all these other alternative [12:49] framework that I that I talked about. [12:52] Um is there questions before we go a [12:54] little bit? This is kind of an overview [12:56] of what ROS is. [12:59] Cool. Okay. So, we're going to go a [13:00] little bit more kind of concrete stuff [13:02] about ROS. So, just real quick history. [13:04] Uh [13:05] so, ROS came out of Stanford, well, in [13:07] part out of Stanford of the [13:09] uh a project on the it's called the the [13:11] STAIR robot. Uh but really kind of the [13:14] people the main people is like a, you [13:16] know, the the magic group of people is [13:18] the Willow Garage group that was uh [13:20] funded by uh Scott Hassan who um did a [13:23] lot of awesome things, but one one [13:25] probably one of the his biggest [13:27] contribution to the field uh was [13:28] definitely to start Willow Garage that [13:31] that put a lot of things out there [13:32] including the PR2 robot which you've [13:34] probably seen before. It's I should have [13:36] included a picture here but it's [13:38] you know this it kind of looks like a [13:39] fridge and it has two arms and it looks [13:41] a little little uh [13:43] uh clunky now but uh it is still is [13:45] incredible back in the days and it's [13:46] still in a pretty good platform. Uh but [13:49] Willow Garage basically was kind of a [13:50] startup but it is almost kind of a [13:52] non-profit startup in the sense of if [13:54] you look at everything that they've done [13:55] for the community in general [13:57] is pretty incredible. And now if there's [14:00] kind of a descendant of what Willow [14:01] Garage was is probably the open source [14:04] robotic foundation which is actually in [14:07] in the Bay Area and that's a non-profit [14:09] that is basically trying to make sure [14:10] that we're funding people that are [14:12] developing for ROS and are developing [14:13] tools for ROS and Gazebo um and things [14:16] like that. So so Willow Garage is [14:18] probably like kind of the you know magic [14:20] word behind ROS if you have to to think [14:22] of one and OSRF is the modern equivalent [14:25] of this. [14:27] Um cool. It also ROS has a lot of [14:31] uh existing like existing project that [14:33] it's integrating. Um so a lot of the [14:36] things that you're probably using now if [14:37] you're working like computer vision for [14:38] example I I suspect a lot of the [14:41] libraries that you're used to if you [14:42] just look for like a ROS wrapper so it's [14:44] a way of integrating whatever libraries [14:46] you're working on like for example [14:48] OpenCV or OpenSLAM if you're doing [14:50] navigation a lot of them are [14:53] people have spent time making sure that [14:54] they are integrated with ROS very [14:55] easily. Uh Okay in terms of actually now [15:00] like if we're looking at what ROS code [15:03] actually is uh [15:05] then it's basically breaks down into [15:06] components. So we had you know we had [15:07] these like these nodes and these topics [15:09] and those are kind of like abstract [15:11] concept. [15:12] This is more like what if you look at [15:14] like the source code of ROS like this is [15:17] what you'll you'll find. [15:18] It's basically two things really. I mean [15:20] the core of it is two things. So, master [15:24] and nodes. So, the nodes are exactly [15:27] what you'd expect from this pub-sub [15:29] architecture. So, they're basically [15:30] these modules that can either publish or [15:32] subscribe to topics. [15:35] And [15:36] and they can they can they talk over [15:38] these topics with these message [15:40] abstractions, which are they can be for [15:41] example they can represent camera images [15:43] or laser scan data or motion control. [15:47] And all the all of the things you're [15:48] going to talk about in in 274 [15:50] essentially probably have like a message [15:52] equivalent in ROS that you can go and [15:55] find like a pose estimate or [15:58] you know a an edge detection image. So, [16:02] Yeah, so we'll do a little bit. I'll [16:04] talk a little bit about master later. [16:06] We'll dive in a little bit into what the [16:08] nodes look like first. So, [16:10] what's what's a node? So, node is is a [16:13] process [16:14] that runs on computation. I guess it's [16:16] like the most general way of calling [16:18] this a piece of software. But [16:21] essentially [clears throat] you can [16:21] think of it like if you're in Python [16:23] it's really just kind of your little [16:24] Python script with your main function, [16:25] right? That's like what like a ROS node [16:27] would look like. It's an individual [16:29] process [16:30] and it will run some algorithm. And and [16:33] most likely it will run some of the [16:34] algorithm that you that you've you'll [16:35] learn in this class. It's the [16:37] fundamental building block. So, [16:38] basically if you have to tell me what [16:40] your robot runs, you're probably going [16:42] to describe it as like what your soft [16:45] the software of your robot looks like, [16:46] you'll probably describe it as just a [16:47] bunch of nodes. And that'll basically be [16:49] enough for me to understand what your [16:51] robot how your robot is architectured. [16:53] Can I again as a subscriber, as a [16:55] publisher? [16:58] Nodes again talk with each other over [17:00] what we call topics, which is again [17:02] that's referencing to this pub-sub [17:03] design [17:04] architecture. [17:06] And the way to run it is there's [17:07] specific commands. So, you as part of [17:10] the workshop classes, so that's I think [17:13] it's going to start tomorrow if I'm [17:14] correct on you going to do the ROS [17:16] workshops on Fridays? [17:18] >> No, on Monday. [17:19] >> Oh, Monday. Okay. [17:20] >> No, the sections. [17:21] >> Right, sections. Sorry. Oh, this class [17:23] is like expanding every year. That's [17:24] awesome. [17:25] Uh so, uh the you're going to you're [17:27] going to install the ROS package. So, [17:29] you'll just [17:30] um you're just going to install the [17:31] software. You know, there's an installer [17:32] online kind of like you have installed [17:33] like Microsoft Word or whatever. [17:35] Uh and then uh with that was going to [17:37] come a bunch of commands that you can [17:38] run. And And one of them is going to be [17:40] for example rosrun. You're going to say [17:42] a package, for example like my [17:43] navigation package, and then a specific [17:45] the name of the node that you want. So, [17:47] let's say like a [17:48] uh lidar uh driver. Like that could be a [17:51] a node. Um and there's a specific [17:53] command to initialize them. We'll go a [17:54] little bit into code later so you get [17:56] like an even more concrete idea of what [17:58] ROS looks like. Uh but is it once you [18:02] understand that like now these nodes uh [18:04] can kind of give you this like nice [18:06] little way of breaking down your system [18:07] into these parts. [18:09] Uh and so and uh that's like one example [18:11] of the type of nodes that you'll have in [18:13] like a system is going to be nodes that [18:15] corresponds to specific sensors. So, [18:16] like I said lidar. So, on the you know, [18:18] on Atlas robot like this which is Oh, [18:20] that's the old version. Um but you know, [18:23] a giant robot it has a ton of sensors, [18:24] right? Uh it has you know, a fish eye [18:27] camera. It has a lidar. It has an IMU. [18:31] Um [18:32] and strain gauges for for pressure in [18:34] the feet. Uh all these things basically [18:36] in the software are can be all the [18:38] individual sensors can be abstracted as [18:40] nodes. Uh and then [18:42] >> [clears throat] [18:42] >> uh the same for actuators. So, a motor [18:44] in Atlas [18:46] uh if you're building your software with [18:48] ROS would probably be a node, like an [18:50] individual node. [18:53] And that's also true. So, So, now uh and [18:56] you know, a sensor can be a node, but [18:57] that's also true of things that are a [18:59] little more abstract. So, algorithms [19:01] could be a node. So, now you'll you've [19:03] learned uh I think you've you've done [19:04] kind of you know, a [19:06] uh a dynamics model for your robot. So, [19:07] you might have a node that is actually [19:09] just really good at at you know, [19:10] predicting what the dynamics of your [19:12] robot is well like simulating forward [19:13] your your robot essentially. And that [19:15] can also be a node. So it's kind of [19:17] weird, right? You have this architecture [19:18] that is just made out of node. Some of [19:20] these nodes are running like a motion [19:22] planning algorithm. That's pretty [19:24] abstract. [19:25] Uh and it's [19:27] and then another node beside that [19:29] that looks almost the same in your [19:30] architecture is actually running like a [19:32] piece of hardware, like a motor that's [19:34] in the physical world. So it's pretty [19:36] awesome. Now you have like this [19:37] architecture that has both parts that [19:39] are abstract and parts that are actually [19:42] like physical real world components that [19:43] are talking to each other. And it's like [19:45] a very easy way to think about uh ROS [19:47] gives you this kind of very easy way to [19:48] think about like organize them together. [19:50] >> Um you mentioned earlier we can think of [19:52] nodes sort of as like uh individual [19:54] scripts. Um but since they're running [19:56] independently of each other, if you have [19:57] one node for dynamics and one for [19:59] sensors, how do they communicate with [20:00] each other and make sure that they still [20:02] run simultaneously? [20:03] >> Yeah, that's a super good question. Um [20:06] so that's where the the pub/sub really [20:08] is what gives you kind of that ability, [20:09] right? So you could think of like the [20:11] the dynamics uh [20:13] or if you have like a simulation [20:15] uh that that simulates forward your [20:17] dynamics of your robot node could just [20:18] be basically running indefinitely, but [20:20] all it needs maybe as input is like your [20:22] current uh your some current sensor [20:24] measurement of like where you are and [20:25] then it rolls out, you know, the future [20:27] prediction. Um so all it really needs to [20:29] know is kind of what it subscribes to [20:32] and then what it needs to publish. So as [20:34] long as that's well established for each [20:36] node, uh you don't necessarily have [20:38] problem in synchronizing. But uh it is [20:40] it it can be difficult to like [20:42] synchronize obviously like sometimes [20:44] you'll be well one node needs to [20:45] communicate to another node that I need [20:47] this piece of information you're not [20:48] sending it. Uh and and this becomes like [20:50] the art of, you know, engineering the [20:52] system. Uh but in general like pub/sub [20:54] has pretty good solutions for that. [20:55] Yeah. [20:56] >> So um is like the frequency of a topic [20:59] limited by [21:00] uh [21:01] nodes or is there like a ROS also on ROS [21:04] side frequency? [21:05] >> Um so you definitely have like [21:08] like your computer can only like run at [21:10] a certain speed. Obviously, that's like [21:11] an bottleneck at some point. But, yeah, [21:13] individual node is basically going to [21:14] decide and you'll see that in the code. [21:16] That's like actually something that you [21:17] set. Even like the most simplest [21:19] node that you you can write that's like [21:21] a something that you can set. So, the [21:23] node just decide like I'm going to [21:24] publish this at like a hertz 1 hertz and [21:26] then and then the other ones can can I [21:28] mean that that might be good or bad [21:30] depending on kind of what you need that [21:31] that node to be doing. Um, but yeah, [21:33] it's basically up to you like as a [21:34] designer. [21:35] >> But, I mean there's ROS throttle limited [21:37] or [21:38] >> Thanks throttle different. Um, [21:43] >> throttle [21:43] >> Yeah, there's definitely yeah, there's a [21:44] point at which point like the I mean [21:46] it's mostly like your networking [21:48] architecture that's not going to be able [21:49] that's going to be a bottleneck, but [21:50] there's nothing as far as I know there's [21:51] no mechanism in ROS that throttles the [21:53] individual nodes. [21:54] Yep. [21:55] >> So, in this schematic, white boxes the [21:58] nodes and the green ones the topics or [22:01] >> Oh, [22:03] Yeah, good question. I'd say this is a [22:06] little bit um, [22:08] Yeah, I don't I wouldn't say this [22:09] corresponds one-on-one to necessarily [22:11] node and topic. Uh, it's more like the [22:13] the white box would be nodes probably if [22:15] you were to like design something. And [22:17] then probably more like the blue arrows [22:19] would be different topics. Uh, but yeah, [22:22] I guess it's a little weird cuz they're [22:23] also linked in the in the green cube. [22:25] But, I wouldn't necessarily read too [22:26] much into this diagram. It's just like [22:28] to say that, you know, individual like [22:29] algorithms are can be nodes and [22:32] individual sensors can also be nodes. [22:33] So, it's just kind of like this this [22:35] very generic thing that that generalizes [22:38] across both software and hardware. [22:40] Um, some specific examples of this, you [22:42] know, like very very concrete, you know, [22:44] you can have a lidar one, you can have a [22:46] lidar node, you can have a node that [22:47] corresponds to a camera that's [22:48] publishing your images. That's that's [22:50] like the topic that's sending out is [22:52] some some image topic. Uh, you can have [22:55] you know, a some IMU that's sending out [22:58] a dometry data. So, that's you know, [22:59] acceleration [23:01] and [23:03] and you know, rate of rotation, [23:05] lidar, so all these things are actually [23:07] things you're going to deal with in this [23:09] class. I think every single one of them. [23:11] Yeah, that's kind of cool. Uh [23:14] and especially lidar cuz that's super [23:15] expensive and then somehow Professor [23:17] Pavone has some that you guys can use. [23:21] Yeah, okay. ROS master. So, that's [23:23] basically it for node. Are there [23:24] questions for nodes? [23:27] We'll go into also like code [23:29] and I go a little bit that might help [23:31] make this more concrete. But, we're just [23:32] going to talk about ROS master a little [23:34] bit. [23:35] So, ROS master doesn't really have an [23:37] equivalent in the in the like if you [23:39] think of this pub-sub design that we're [23:41] talking about where we have these nodes [23:43] and we have these topics, then ROS [23:44] master is not really like a part of that [23:46] architecture. It's just kind of a like [23:48] necessary evil that a little bit of like [23:50] once you actually start implementing [23:51] this, someone has to manage the forum, [23:54] someone has to manage the Slack channels [23:55] basically or the mailing list. Uh so, [23:58] you can think of that that's like that's [23:59] what ROS master is. It's really just [24:01] this process that basically you have to [24:02] start every time you start your stack. [24:05] And what it'll do is it'll it'll [24:07] it'll coordinate the nodes. It'll [24:09] coordinate the publisher and the [24:10] subscriber. [24:11] You always only need one on a network. [24:13] So, as long as like you want to every [24:15] node to talk to each other, one ROS [24:17] master is enough. And something to to uh [24:22] keep in mind so that's that that becomes [24:23] like a pretty common pitfall in the way [24:25] that I've had like very very [24:27] um experienced robotics you know, make [24:30] that mistake to think that the messages [24:33] go through the ROS master. They don't. [24:36] Uh and that that's like important. [24:38] Basically like what ROS master is going [24:39] to do is going to set up the topic. So, [24:41] if you have a node that's subscribing [24:42] and a node is publishing, it is going to [24:44] set them up on some topic. [24:46] Uh and if you want to you know the know [24:48] the details that's like it's really just [24:49] putting like a TCP connections between [24:51] the two, but it's going to connect them [24:53] and then it's basically going to back [24:54] off. So, your ROS master yeah, that that [24:57] will I promise you, this will like come [24:58] up in your final project. You'll be [25:00] like, "Oh, it's cuz my master is like [25:01] the ball neck or something." Like, all [25:03] the networking is trying to go through [25:04] ROS master and ROS master is too slow. [25:06] And it it's not what's happening. Like, [25:08] ROS master is just like setting up [25:09] connections and then backing off. [25:12] Uh and then letting the nodes talk to [25:13] each other. [25:15] >> [clears throat] [25:15] >> Yeah. [25:16] >> Does it matter if you start the ROS [25:17] master on a different computer or if you [25:19] off computer? [25:20] >> Yeah, good question. Um [25:22] yeah, not really. [25:23] >> Okay. [25:23] >> You could do either. Uh [25:25] like, I mean, the [25:26] the only thing maybe that matters is [25:28] that uh [25:28] in reality your robot's probably going [25:30] to crash a bunch. Um so, it can be nice [25:33] to not have ROS master on the robot just [25:35] so that that way it's always kind of [25:36] running somewhere and you don't have to [25:38] like restart it and everything. Uh but, [25:39] that's a good very like in theory it [25:41] doesn't matter in practice because [25:42] robots crash all the time. Uh like, you [25:44] might want to put it somewhere that's [25:46] like safe. Uh [25:47] yeah. [25:49] >> All right, if it's just for initially [25:51] connecting all the nodes, then why is it [25:52] a process? Like, can't it just be an [25:54] initialization sequence and then it's [25:56] just done? [25:57] >> Yeah. Um yeah, yeah, great question. Uh [26:00] in theory you could. Uh the the thing is [26:02] a lot of these [26:04] um a lot a lot of these systems have [26:06] nodes that will like appear and [26:07] disappear [26:08] at run time. So, as you're running your [26:09] robot, like it might like, "Oh, actually [26:11] I'm in this new mode of operation. Uh [26:14] so, like maybe now I'm running, you [26:15] know, I'm at like a I'm at a I'm like a [26:17] self-driving car and I'm at a stop sign. [26:20] Uh and now I want to start a node that's [26:21] actually like tracking pedestrian." And [26:23] that was not something I was doing [26:24] before cuz I was on the highway. Uh and [26:26] so, you're bringing up this node and [26:27] then that node needs to make sure that [26:28] it's telling your path planner that [26:30] there are pedestrians around. Um so, for [26:32] these types of reason that like nodes [26:34] come in and out of the system online a [26:36] whole bunch, uh [26:38] in general like you'd want to have I [26:40] mean, it won't ROS won't even let you [26:41] run without ROS master, so you'll you'll [26:43] need it. Um [26:45] but but that's like that's the the [26:47] practical reason for that. Uh just they [26:49] the nodes kind of appear and disappear [26:50] all the time. [26:53] Is that good? [26:54] Cool. [26:56] Um okay, let's look at code. [26:58] Code's fun. Um yeah, so this is a little [27:01] bit like just you know the lines you're [27:03] going to write this a whole bunch over [27:04] the quarter. Uh basically that's what it [27:06] looks like if you're in a node which and [27:08] this class you guys are going to use [27:09] Python. Um [27:11] you but uh just for the record uh ROS [27:14] also has you know C++ libraries. Uh it [27:16] even has you know things like Julia [27:18] libraries which uh I encourage you to [27:20] look up. Uh but the Python uh [27:23] the Python [clears throat] interface [27:24] looks something like this. So on your [27:25] Python script, you know, sending a [27:27] message is as simple as uh making sure [27:30] that you're creating a publisher [27:32] and then uh you create a message which [27:34] is basically going to look something [27:35] like you know just creating like a [27:36] Python object and putting like a string [27:38] in it essentially. Uh and then you're [27:40] going to use your publisher to publish [27:41] the message. And then that's it. It's [27:43] out on your topic and whoever is [27:45] listening for for your trajectory that [27:48] you've just generated is can receive it. [27:51] Um [27:52] cool. Let's look a little bit at what [27:54] the talker would look like. Um [27:57] so actually yeah, let's look over to the [28:00] publisher and the subscriber first. So [28:01] just going to go a little bit line by [28:03] line on a publisher and subscriber and [28:04] then we're going to actually run it. Um [28:07] uh on on the soft top. So essentially [28:09] it's just Python script. I'm sure I'm [28:11] pretty sure Python is a pre-req for this [28:14] class. So yeah, I can see you guys are [28:17] giving me a thumbs up. Uh so you guys [28:19] should all be able to also read this. Um [28:21] but [clears throat] [28:22] uh this is just a generic Python script [28:23] that you guys [28:24] um recognize and then so all all this is [28:27] doing essentially uh you have you know [28:29] your main function uh and then it's [28:30] talking talker uh and then your talker [28:33] now is a node that is we're going to [28:35] make it explicitly a publisher. We could [28:36] also make it a subscriber, but this [28:38] particular node is going to be a [28:40] publisher only um just for the sake of [28:43] making this simple. So what this looks [28:45] like. So the first thing we're going to [28:46] do is just we're going to see that like [28:48] hey, this Python script is [28:50] is a node a ROS node. So, that's that [28:52] basically what that first line is doing [28:54] is basically just saying I'm a ROS node, [28:56] I'm called talker. Um, you probably want [28:58] to come up with better more more [28:59] descriptive names than that when you're [29:00] actually going to implement this. [29:03] The anonymous keyword is basically [29:06] is just basically that so that you could [29:07] have multiple [29:09] nodes with that name. [29:11] ROS is going to make sure that basically [29:12] like that doesn't conflict. Cuz [29:14] otherwise each node has to have a [29:15] different name if you set anonymous with [29:18] false. You're going to create a [29:18] publisher. [29:20] The first argument here is basically [29:21] just saying what the topic is. So, you [29:24] know, that big like blue colon [29:25] essentially like that has, you know, [29:26] that's like the topic. That's like the [29:28] Slack channel or the mailing list. So, [29:30] in this case we're just going to be a [29:31] topic chatter. It's going to be super [29:33] interesting. [29:34] And then the message type, so in that [29:36] case it's a string. Something to notice, [29:38] this is not a Python string object. It's [29:41] a ROS message string. So, at the top we [29:44] we actually imported something from the [29:45] standard message library which is [29:47] something that that comes with ROS. Um, [29:49] and so it's basically saying okay, we're [29:50] going to submit over chatter. We're [29:52] going to submit we're going to be [29:53] sending string type messages. And then [29:56] we have a queue size which is basically [29:58] just saying like [30:00] that if for whatever [clears throat] [30:01] reason for networking reason which is [30:02] you know, to your point earlier if you [30:03] ever get throttled, if you're kind of [30:05] trying to send things too fast, that [30:07] node is going to make sure that like [30:08] okay, I might drop like the old messages [30:10] messages and like at what point do I [30:11] start dropping messages or at what point [30:13] do I [30:14] like how many do I keep you know, if I [30:16] get a bottleneck from the network. Um, [30:19] then it's going to get a param which is [30:20] a parameter which is the rate at which [30:22] it's going to publish. So, again to your [30:24] point earlier that like that's actually [30:25] a parameter. Then in that case that we [30:27] made it that you can set it from [30:28] somewhere else in your system, but that [30:30] could also just be hard coded. Just [30:31] basically saying ROS is going to run at [30:33] this rate. [30:34] This also maybe some of you would be [30:36] tempted to implement this you know, like [30:37] in terms of like oh, I want to run this [30:39] loop and [30:41] I mean you guys have done plenty of [30:43] coding. You might be like, "Okay, I'm [30:44] going to import like time and I'm going [30:45] to, you know, have like a timer that [30:47] starts and stops." You don't want to do [30:48] that. And the reason for that is the [30:50] reason we want to use the raw straight [30:52] is because you might actually connect [30:53] this to for example a simulator. Uh and [30:56] your simulator might not be running in [30:57] real time. So, it's a bit of like a [31:00] subtle thing, but essentially like your [31:01] simulator might be running like slower [31:03] than real time or even faster. [31:06] Uh and if that's the case, then you want [31:07] to make sure that all the other nodes [31:09] that are like talking to your simulator [31:10] are using the same like time. [31:13] Uh so, that's why ROS is kind of using [31:14] its own clock essentially and that clock [31:16] usually is, you know, the same as the [31:18] clock on your wall, but um for some [31:20] specific applications sometimes you want [31:22] to make it run slower or faster in real [31:25] time. Yep. [31:27] >> What's the topic? [31:29] >> Uh [31:30] do you mean like which argument was it [31:31] the topic? So, in that case it would be [31:34] uh chatter. [31:35] So, yeah, the first like argument of [31:37] when you submit then you create a [31:38] publisher is like what topic is going to [31:40] publish on. [31:41] Um [31:42] and then we're just going to log some [31:43] information. That's like essentially the [31:45] equivalent of like your print line [31:46] statement, right? Which you've probably [31:47] done a whole bunch of. [31:49] Uh and then all this is doing is going [31:51] to put the message which in our case is [31:53] also just a string. [31:55] Um sometimes you might like have to like [31:57] this message might actually be an object [31:58] like a you know a dictionary, right? [32:00] Which has like your IMU measurement in [32:01] it, like your accelerometer measurement, [32:03] your gyro measurement, and then you'll [32:04] have to like fill all of these up. And [32:06] then all you do is you publish it and [32:07] then you wait for whatever amount of [32:10] time that you need to wait and then you [32:11] just repeat. And that's it. That's like [32:12] anatomy of like a super simple node. [32:15] Yep. [32:15] >> So, is it standard to sort of construct [32:17] these nodes just with the function as [32:19] opposed to like actually creating an [32:20] object that's of a certain class and [32:22] giving it attributes? [32:23] >> Um [32:24] this is like as simple as you can do it. [32:28] I would say that in practice, yeah, [32:29] probably like [32:31] I don't know. I mean that's like very [32:32] much like a I guess a style thing, but [32:34] like 80% of the time I I have some sort [32:35] of class wrapper. [32:37] Um [32:38] some of the things that might happen is [32:39] like for example like one one of these [32:41] class might actually contain like a both [32:43] a publisher and a subscriber. So it's [32:44] one node but it can both publish and [32:46] subscribe. And so like you might be [32:47] subscribing to a topic, you want to [32:49] store that in that class, and then maybe [32:52] you're running some path planning [32:53] algorithm that's using this data and [32:54] then publishing it again at some rate. [32:56] So it's actually yeah pretty common to [32:58] like have some some class that wraps [33:00] things. [33:01] But yeah it's a it's a good point. It's [33:02] a good way to organize your code too in [33:04] general. But like for this it's just to [33:06] make it simple basically. Yeah. [33:09] >> We added ros rate dot sleep is like [33:11] common practice or do we typically use [33:13] like state machines to make sure that [33:14] it's constantly checking for interrupts [33:16] and stuff? [33:16] >> Um [33:17] >> [clears throat] [33:18] >> Yeah so if you're it depends what you're [33:20] writing. [33:21] If you're doing anything that has to run [33:23] at a certain rate then yeah doing this [33:25] kind of like that this ros rate thing is [33:28] probably the way to go. [33:31] Sometimes so you'll see on the [33:32] subscriber that I'll show I'll show next [33:34] that's maybe more what kind of you have [33:36] in mind where you'd have like an [33:37] interrupt. Like you might want to just [33:38] do something when you get a message. [33:41] And then and then then that's like [33:43] probably more like a subscriber. But [33:45] yeah it depends what you're doing [33:47] basically. [33:48] I would say that's probably the most [33:49] common version but it's an excellent [33:51] point that some stuff you might not want [33:53] to run at 10 hertz. Some stuff you might [33:54] just want to run whenever you get an [33:55] image in you want to run some some like [33:58] computer vision on it. You will see in [34:00] the stack that you use like the for the [34:02] final project [34:03] you'll have both of these things. Like [34:05] they'll be all over the place. [34:08] Cool. Um so okay we have this node [34:11] that's publishing and uh [34:14] uh [34:15] there's a you know now you might want to [34:17] know what it's publishing at what rate [34:18] it's publishing. And so there's a bunch [34:19] of tools that come with ROS to give to [34:21] help you do that. So ros topic list, ros [34:23] topic echo if you give the name of the [34:24] topic will print it. Ros topic hertz [34:27] super useful for your final project cuz [34:29] you'll be like why is my state [34:30] estimation all wrong? And then you'll [34:32] find out that your lidar is sending you [34:34] measurement at like point 01 hertz. So [34:37] you get one every What is that? 10 [34:38] seconds? Or no, 100 seconds? So, yeah, [34:41] that's probably why. [34:43] And so, yeah, all of these are useful. [34:44] Not you probably won't remember them [34:46] now, but I'm just putting them here in [34:47] the slides so you have them as a [34:48] reference. There was a question in the [34:50] back. Go ahead. Sure. [34:52] >> On the previous slide, where exactly was [34:54] the master being involved in that [34:56] process? [34:57] >> Nowhere here. [34:59] Well, [35:00] okay, so you'd start separately. This [35:02] here, like the init node, is what like [35:05] and like at the back is probably is [35:07] talking to ROS master and like setting [35:09] up the setting itself [35:11] like registering itself as a node of the [35:13] system. [35:14] But there's not nowhere here that like [35:15] you're you're running master. [35:18] I'll I'll show you the demo, but you'll [35:19] see like you start master separately and [35:20] then you start your nodes essentially [35:21] like the process. And then and they they [35:24] like register themselves with master on [35:25] that first [35:27] Those are at the back. [35:28] >> Is it possible to go over the [35:30] >> [laughter] [35:30] >> difference of defining rates and two [35:32] arguments that go into the get parameter [35:34] function in the definition of the rate? [35:38] >> Yes. [35:39] You mean this? Okay. So, to be clear [35:42] like this could be [35:45] Okay, so [35:47] this is this is basically just to say [35:48] you want to use like a ROS rate object. [35:50] So that's like the what I was talking [35:51] about like you might not want to use [35:52] like a real [35:54] real clock. You want to make sure that [35:55] like you're saying I'm using the ROS [35:57] clock. So that's why this is a different [35:59] rate object. The top one essentially [36:02] what it's saying is like look up for a [36:03] parameter that's called rate. So you [36:06] have like a central like ROS master also [36:08] does a whole bunch of other little tasks [36:10] to coordinate everyone. One of the thing [36:11] it does is store parameters [36:14] for your system. And so one of those [36:15] parameters that you could store is like [36:17] rate. [36:18] And so what this is is if you've used [36:19] dictionaries before, [36:21] you know, this is literally like looking [36:22] up a dictionary and telling you if you [36:24] don't find that entry, [36:26] uh back on this default value. [36:29] Yeah. [36:31] Uh-huh. [36:32] >> What's anonymous? [36:34] >> Uh so, that's uh what I was what I was [36:36] saying earlier. Uh basically, if you [36:38] have different node if you have [36:39] different node that have uh the same [36:41] name, but they're created as anonymous, [36:44] um [36:45] then ROS is going to be okay with it [36:46] because it basically it'll add like uh [36:48] unique ID to the end of each name of [36:51] each node, and so they can run like with [36:53] all the same name. [36:54] Uh but otherwise, if you if you set [36:56] anonymous to false, ROS is going to the [36:58] ROS master is going to make sure that [36:59] like no node has the same name. Uh and [37:01] it'll like basically, if you create a [37:03] second node that has the same name as a [37:04] node that already exists, it will uh [37:06] kill the like older one. Um yeah. [37:10] Cool. Okay, if you have any questions [37:11] also about the syntax, [37:13] uh we should definitely like answer some [37:15] of these questions, but also like don't [37:16] don't like panic. Like all of this is [37:19] super well documented. Um I think I just [37:22] want to give you like an overview and [37:23] everything, but I promise you this you [37:25] know, this piece of code is like [37:26] probably like the first thing that pops [37:27] up on Google. So, it's like very easy to [37:29] to find, and that's why like we have ROS [37:31] in this class is because all of this [37:33] syntax is like really easy to find [37:35] online. Um cool. [37:38] >> [clears throat] [37:38] >> Okay, receiving message. Uh we're just [37:40] going to go through this real quick, but [37:41] basically it's pretty similar uh except [37:43] we're going to use something called a [37:44] callback for this. So, when you're going [37:47] to create a subscriber, you're going to [37:48] say uh you're going to give it uh a [37:51] uh you're going to give it uh a topic [37:53] name that it's subscribing to. So, in [37:54] our case, chatter, right? If you [37:56] remember, there was like the first [37:57] argument of that that other function [37:59] before where we were saying we're going [38:00] to talk on the chatter topic. Uh now, [38:03] this is the one that we're going to [38:03] subscribe to. Uh we're going to say that [38:05] we're expecting string uh type messages, [38:09] and we're going to tell it, "Okay, once [38:11] your subscriber, once you get a message [38:13] of this type on this topic, this is the [38:15] function that you should call with that [38:17] message." So, that's the callback uh [38:19] here. And then what you have to do as [38:21] like the implementer of this is to [38:23] implement the callback. So, in a more [38:24] like [38:26] concrete way what this looks like, um, a [38:29] little a little smaller, but actually [38:30] kind of more subtle. Like I think I [38:32] think the [38:33] the subscriber is actually a little more [38:34] complicated to understand. [38:36] Um, [38:37] so we'll go through this as well. So, [38:38] again, uh, so all this is doing [38:40] basically is it's starting this this [38:42] listener uh, topic. [38:44] Uh, [38:45] which is essentially a for In this case [38:47] you're not going to have to have a for [38:48] loop. Uh, this is what kind of this raw [38:50] spider spin is going to do for you. [38:52] Essentially it's going to implement this [38:54] like this like [38:55] continuous for loop at like that fixed [38:57] rate that you had. You don't have to do [38:59] that. Uh, you know, to your point [39:00] earlier, do you always have to have a [39:01] rate? If you're running a subscriber, [39:03] maybe not. Maybe all you want to do is [39:04] do something every time you get a [39:05] message. Uh, so that's is what this [39:07] would do. So, you'd in it you'd [39:08] initialize a node, right? That's no [39:10] different than our publisher because [39:12] nodes are all the same and they all live [39:14] in an equal world where they just have a [39:16] name and they're anonymous or not. [39:18] Um, they're going to you're going to [39:19] define our subscriber. You're going to [39:21] say, "Okay, this node has a this node is [39:23] subscribed to a topic. Uh, in this case [39:25] chatter." Again, like we said, [39:27] the message type and what it's going to [39:29] do once it gets one. And this is just [39:30] like a print line. Uh, [39:32] and now, [39:33] uh, raw spider spin is basically going [39:35] to say, "Okay, now just uh, [39:37] send stuff." And and then if you ever [39:39] get any messages on any of your [39:41] subscriber that you've created, call the [39:43] associated callback. So, the callback [39:45] here, which is, you know, I guess called [39:47] callback. It's a callback that's called [39:49] a callback, which is probably a little [39:51] annoying. Don't do that. That's going to [39:52] be confusing for you. But, uh, all it's [39:55] doing is like this message this callback [39:57] just has to take a message and when it [39:58] gets it, it prints it. Uh, that's it. No [40:01] message contains like message of data [40:02] that will just recover like the string [40:04] that you had. [40:05] Yeah. [40:07] >> Oh, creates a message out of raw spider [40:09] right from the callback and processes [40:10] it. Does it add to that queue? [40:12] >> Uh, [40:13] yes, on the queue of the publisher. You [40:14] can You can write queues for the [40:16] subscriber. [40:17] Um which I think is just like a like a [40:20] not mandatory [40:22] uh [40:23] uh argument or yeah, optional argument. [40:26] That's what I'll go for. Yeah. Um so [40:28] yeah, there's a bunch of like queuing [40:29] you can do between everything, uh but [40:30] that's all like [40:32] that's not like usually needed, uh but [40:34] it's probably not going to be needed for [40:36] you guys unless you're doing like super [40:37] high rate things. [40:39] Uh-huh. [40:41] >> Messages if you're not explicitly [40:42] defining a rate in which the the [40:44] listener kind of listens to the to the [40:46] messages that's going to be queued. [40:49] >> Why would there be a queue? [40:50] >> Yes, if you're not explicitly [40:51] >> Yeah, so if your callback is taking more [40:54] a lot of time, [40:55] um so you know, in this case it's just [40:57] like printing a message every time, [40:58] right? But the message you might get [41:00] might be like, "Hey, solve this [41:01] optimization problem." And maybe this [41:03] optimization problem takes like 10 [41:04] minutes. Uh and then maybe like the [41:06] whoever's sending you these problems is [41:07] like sending them to you at like every 5 [41:09] minutes. Uh so you're kind of stuck, [41:11] right? Now you're solving these problems [41:12] and you're just like kind of piling them [41:13] up uh and and like not solving them in [41:15] time. [41:16] Uh and yeah, these problems come up all [41:18] the time. That's like definitely [41:19] something that you have to watch out for [41:21] if like who's keeping up with what [41:23] uh is the thing to monitor. But yeah, [41:24] good question. Yep. [41:27] >> register with the master because I don't [41:29] see it show up anywhere afterwards. [41:31] >> Correct. Yeah, you basically just have [41:33] to do that once. Um they register it as [41:35] like, "I am a node in the system. Listen [41:38] to me if I'm trying to subscribe to [41:39] things or publish things." But yeah, [41:41] that's what that's like really what's [41:43] differentiating this from like a regular [41:45] Python script is really just making sure [41:46] that like you're calling this like [41:47] library thing, this library function [41:49] that subscribes you. That that tells [41:51] master that you exist as a node. [41:54] Cool. Yep. [41:55] >> So if you have a publisher that's [41:56] publishing way faster than your [41:58] subscriber [41:59] >> Uh-huh. [41:59] >> and you're publishing like 700 or [42:01] something [42:02] >> Uh-huh. [42:02] >> you set it so that you can look at just [42:04] the most recent thing. [42:05] >> Yeah. [42:05] >> Okay. [42:06] >> Yeah, no, absolutely. You can have like, [42:07] you know, you can make it drop a bunch [42:08] of like the latest messages or drop like [42:10] the oldest ones or all this queuing [42:12] stuff. Um I'm going to give a quick note [42:14] on like raw stew at the end. Uh but like [42:16] people are getting pretty fancy about [42:18] the way that they're handling this cuz [42:19] it matters. [42:20] Yeah, but hopefully you design fast [42:22] algorithms uh the way Professor Pipitone [42:25] is going to teach you to and that never [42:27] happens. [42:28] Uh okay, we're just going to run this to [42:30] make this like very concrete. Uh so [42:34] You guys see all of this? Yeah, we'll [42:35] make this bigger. [42:41] Okay. [42:42] Um so the first thing we're going to do [42:43] is we're just going to run raw score. [42:46] >> [clears throat] [42:46] >> Is that big enough or [42:48] for people to see? [42:52] I'll take that as a yes. [42:57] Okay. [42:58] Um [42:59] Yeah, so this is this is it. Like this [43:01] just run like raw score is master [43:05] uh or like raw master. They're like the [43:07] same thing essentially. So raw score uh [43:09] is just like the one command. You launch [43:11] this. This usually [43:13] lives somewhere on your system. Here [43:15] it's telling you for example [43:17] uh the parameters that it has. Uh so as [43:19] you can see we haven't defined a rate [43:20] parameter. So if I were to initialize [43:22] this [43:23] um this note that we went through [43:25] earlier, uh it would fall back on the [43:28] default that we defined like that second [43:29] argument. Um because the only parameters [43:31] that it has is telling you which raw [43:33] distribution you have. So that's like [43:34] its version essentially. Uh it's using [43:36] kinetic which the TA confirmed me that [43:38] this is all that you guys are using. [43:40] Uh and [43:42] yeah, it's giving you a bunch of other [43:43] information which is going to make more [43:44] sense once you actually start using it. [43:46] But essentially most of the time you're [43:48] just going to be running this and you're [43:49] going to make sure that you know some of [43:50] these parameters kind of make sense to [43:51] you. [43:52] And then the second thing we're going to [43:55] do [43:56] Now we're going to go in [44:04] Okay, so we have our listener and our [44:07] talker, which are literally like the [44:09] code that [44:10] uh [44:11] you know, we just went through. [44:13] Cuz it's it's like word for word. Um [44:15] so we're [clears throat] going to start [44:16] our [44:18] uh our talker. [44:21] And so remember we had the still like [44:23] print statement that was just logging [44:25] information. Um and now it's just [44:27] talking, it's just sending these [44:28] messages at this 1 Hz rate. Uh before [44:32] starting the subscriber, we're actually [44:33] just going to use a tool that comes with [44:35] ROS. So [44:37] um we're [clears throat] going to run [44:38] this thing called ROS topic list, which [44:41] is going to give us the list of things [44:42] that are being talked about, the list of [44:44] topics that exist right now. So chatter, [44:46] right, is the one we created earlier. Um [44:49] there's also always ROS out and ROS out [44:52] agg, which just ignore for now. Uh but [44:54] chatter is like the one that we care [44:55] about. Uh we can even say, "Okay, at [44:58] what rate is this thing publishing at?" [45:03] And it will tell us that, "Okay, it has [45:06] rate 1 Hz." Um it even gives you a nice [45:09] little statistics on the standard [45:10] deviation and everything. So that's [45:12] cool. So this is seems to be working. Uh [45:14] and now we can run We'll run our [45:16] listener [45:17] um our listener [45:19] node. [45:26] So as you can see, we're getting the [45:27] message that our other node was sending [45:30] uh [45:31] and it with the right message content [45:33] content at about like 1 Hz as we wanted. [45:36] Yep. [45:38] >> Do you have to run ROS run or can you [45:40] just run the script directly in the ROS [45:42] run syntax? [45:44] >> Yeah. Uh good question. For Python, you [45:46] can do either. [45:48] Um [45:48] Yeah, it's a little like more [45:50] complicated depending on which language [45:51] you're using. [45:53] Uh I Yeah, I didn't talk about ROS run [45:56] uh too much details in the Yeah, but in [45:58] case of Python script, you can do it [46:00] that way. One thing you will notice uh [46:03] that's like you will probably I don't [46:05] know if you remember this but this is [46:06] like an important little detail maybe. [46:08] If you do a rossrun like this, your [46:09] output [46:11] isn't necessarily being captured. So, if [46:13] you're doing like a print statement in [46:15] your script, like this is probably it's [46:17] probably going to show up. It should [46:19] show up. If you're doing it through [46:20] rossrun, [46:22] uh [46:23] it won't. Like like all the output gets [46:25] captured by ROS essentially. So, you [46:27] have then to use like the actual logging [46:29] uh method. But, this is probably way [46:31] more in the weeds than you need to like [46:33] think about right now. Yeah. [46:35] >> So, the time step, why is it starting [46:36] with like a lot, you know, 15 [46:39] >> Yeah, this is like CPU time [46:42] um [46:42] which is they just have like this [46:44] notation that, you know, it's like epoch [46:46] which I think is start epoch or [46:48] something. I'm not a native speaker. [46:49] But, [46:49] >> [laughter] [46:50] >> uh it starts at like some date in 1970 [46:53] and it's counting like seconds since [46:54] then or something like that. Uh [46:57] It probably is the same. I don't know [46:58] what GPS time is, but it probably is [47:00] similar. [47:01] Uh yeah. But, on CPU usually if you ask [47:03] like a time, there's a there's a number [47:05] that, you know, everyone just agreed [47:07] that like time for computers start like [47:09] January 1st, 1970 and that's like what [47:12] this corresponds to. But, it doesn't [47:13] really matter. Like, you never really [47:14] care whether your robot is [47:16] knows that it's like Tuesday morning. [47:18] What you care about is that it knows [47:19] that it's 5 seconds after the last lidar [47:22] measurement. So, most of the time, you [47:24] know, when you're looking at these time [47:25] steps, you're looking at like relative [47:26] time. [47:28] Cool. [47:29] So, does this demo kind of make sense? [47:32] This sounds like so simple and silly, [47:34] but like this is it. Like, this is how [47:36] you build really, really complicated [47:38] things. [47:39] It's just like having these two nodes. [47:41] And now, you know, piece at one is like [47:43] one node and then like piece at two is [47:44] like the other node and they can talk to [47:45] each other and do really incredible [47:47] things that make robotics very magical. [47:50] But, yeah, under the hood like this is [47:51] it. It's just like passing messages back [47:53] and forth in a very efficient way and in [47:55] a way that like makes it understandable [47:58] for like you the human who's actually [48:00] building this. [48:02] Cool. All right, we're going to move on [48:04] a little bit. [48:08] Um cool. Okay, to your question at the [48:11] back about whether you can use rosrun or [48:14] just run the script. There's even [48:16] another way that you can run this stuff. [48:18] Uh [48:18] so yeah, you will get lost in the other [48:20] ways that you can start your system. And [48:21] that's that's fine. Um but there are [48:23] some ways that at some point, you know, [48:24] you get so many nodes in your system uh [48:26] that you need to start, you know, you [48:28] have this like talker here and this [48:29] listener and then you have maybe like 12 [48:30] other listeners that you need to start [48:32] every time. So, it's kind of annoying if [48:33] you have everyone that has to start, you [48:35] know, their own little their own little [48:37] uh node. So, ROS the ROS community [48:39] developed the a lot of tools you know, [48:40] help you deal with that complexity. Uh [48:42] and one of them is these launch files. [48:43] So, that's also something you'll deal a [48:44] lot with in this class. It's basically [48:46] this XML file uh basically like text, [48:49] you know, just like very structured text [48:51] uh that that tells you that will say [48:53] that will basically tell you how to like [48:55] boot up your entire robotic stack. So, [48:57] we'll, you know, have a list of nodes [48:59] that you want to start. Uh you can set [49:01] some parameters in there. Uh you can set [49:03] some uh a whole bunch of, you know, [49:05] configuration things and rates of things [49:07] uh and it all now comes on in one big uh [49:11] XML file. And that's super useful [49:12] because then you can, you know, once [49:14] you're going to build your final project [49:16] and have all this all these different [49:18] things that you've implemented uh [49:20] ideally, at the end of the day, you'll [49:21] just be doing, you know, roslaunch on [49:23] the robot uh and then you say, "Look at [49:25] this file. It tells you every node that [49:27] you have to start." and it starts them [49:28] for you. [49:30] So, useful thing and uh I'll show you a [49:32] little bit [49:33] uh [49:34] kind of the syntax of what it looks [49:35] like. So, this is like the simplest, you [49:37] know, launch file that you could think [49:38] of. It's just like an XML file. It tells [49:40] you the name of the node uh that you're [49:43] going to start. You're telling it which [49:45] package to find it. We're going to talk [49:46] about package a little bit later. [49:48] Uh the type of node, so that's where you [49:50] would like the actual script um [49:53] that that starts. [49:54] The output, so just saying like that's [49:57] where [49:58] you know, I was saying like the output [49:59] gets captured or not. You can tell it, [50:00] you know, print on the screen or you [50:01] could tell it like print somewhere else. [50:04] And then you can set some parameters, [50:05] right? So, here we're actually setting [50:07] our rate parameter to a value of five. [50:10] So, if we were to run our talker [50:13] through this launch file and not through [50:15] just running Python like we did before, [50:17] the parameter would get set through the [50:19] launch file [50:20] to be five and now our talker would run [50:22] at five hertz instead of the one hertz [50:24] default that it had. [50:26] Cool. [50:28] Okay. Yeah. We'll do We'll do a little [50:30] like more fun example. [50:33] I think the talker listener is like [50:34] awesome, but I can appreciate that maybe [50:36] it's not as awesome for everyone. [50:39] But, let's look at like a slightly more [50:41] complicated system. So, in this case, [50:44] we're going to look at like a case where [50:46] you're trying to do edge detection for [50:48] for camera. So, that's actually an [50:49] algorithm that you'll learn in this [50:51] class that you'll figure out how to do [50:52] some some edge detections [50:55] in maps I think for lidar if that's [50:57] still happening. There's a lot of things [50:59] happening, so [51:00] But, yeah. So, we're going to look at [51:02] like what that would look like, you [51:03] know, if you were to use instead of your [51:04] own algorithm, you want to build this [51:05] with, you know, tools that already [51:06] exist. Like I said, a lot of this have [51:08] already been implemented. [51:09] A lot of really good implementation [51:11] exists for the ROS [51:12] environment and so it's it makes sense [51:14] to try to reuse them. So, in our case, [51:16] we're going to define four nodes. One of [51:18] them is going to be a camera driver. [51:20] Right? So, remember that like picture of [51:21] Atlas with all the sensors and I was [51:22] telling you like some of these sensors [51:24] can just be nodes. We're going to do [51:25] this here where our camera, so my webcam [51:28] on this laptop is going to be a node. [51:32] And then we're also going to do We're [51:33] going to have an edge detection node. [51:35] So, that's more like the second graph, [51:36] right? With the green boxes and the like [51:39] blue arrows where, you know, we're [51:41] saying that [51:42] like a piece of software can also be a [51:44] node. So, that's what this edge [51:45] detection is. It's really slowly just [51:46] like some computer vision [51:48] software algorithm that like you [51:50] probably learned in in computer vision [51:51] class or you're going to learn here. And [51:52] then we're going to have two other nodes [51:54] um which those are actually going to be [51:56] gooey. So, one of them is going to show [51:58] us what's the camera seeing. [52:01] Uh so, it's going to [52:02] and then the other one is going to show [52:03] us what is the edge detection uh produce [52:06] algorithm producing. So, there's a bunch [52:08] of topics [52:10] right? That connects those. And that's [52:12] actually So, if we launch all of them, [52:14] uh you kind of build this in your head [52:16] and then you call this uh [52:17] this utility [clears throat] [52:19] uh little piece of code that's called [52:21] RQT graph. ROS can actually generate [52:23] this graph of what your nodes look like [52:25] and what your topics look like. So, in [52:27] this case, this is the camera driver [52:30] node, right? Um and what this is [52:33] actually doing is it's publishing over [52:35] this [52:36] uh this this publishing over different [52:39] over this topic image raw. Uh so, USB [52:41] cam image raw that is being subscribed [52:43] to in image view. [52:45] Um you can kind of ignore this part. [52:47] It's like internally converting [52:48] something. But essentially, this node is [52:51] is uh subscribing to, you know, image [52:52] raw that's coming from this node and [52:55] just showing it to us. And then this [52:57] other one, the edge detection, is [52:59] subscribing to the exact same topic, [53:01] right? They're both listening to the [53:02] same topic. Uh but in this case, it's [53:04] actually running an algorithm and then [53:06] which is detecting edges uh in the image [53:08] and then publishing on another topic [53:10] called edge detection image to another [53:13] node of our fourth node that is just [53:15] like showing that image with edge [53:17] detected. Does that architecture kind of [53:19] make sense? [53:21] Cool. Yeah. So, like this is you're [53:24] going to do a ton of those. Like this is [53:25] going to be like your sensors, could be [53:27] your IMU data, and then this is going to [53:29] be like maybe a little visualizer you're [53:30] going to build, and this is going to be [53:31] like your state estimator, and then [53:33] that's going to go to like your path [53:34] planner. Um [53:36] So, yeah. This is very much like a [53:38] condensed version of [53:40] of uh [53:41] >> [clears throat] [53:42] >> what a lot of what you guys are going to [53:43] do. Uh I'm actually going to run it [53:45] here. [53:47] So we still have our ROS core running, [53:50] right? [53:51] I'm going to like mix it up a little [53:53] bit. [53:54] What I'm going to do is I'm going to run [53:55] this node [53:56] which comes from [53:58] I should also do this in full screen. [54:02] Okay, can you guys still see at the [54:03] bottom? [54:04] I will read it out loud. So rosrun USB [54:07] cam which is basically saying this [54:09] package that you know you can just find [54:11] online. [54:13] You know that I downloaded yesterday and [54:14] it you know just gives you like a driver [54:16] for your camera. And then which node it [54:18] has the USB cam node. [54:20] So all this is going to do is run like [54:22] like one that first node, right? The [54:25] driver node. [54:26] And so we're using [54:28] And that's a problem. [54:39] Okay, we have a bit of a bit of an [54:41] issue. Uh This is interesting. [54:50] Okay, I will we'll move on and try to [54:53] get back to it. It looks like I've had a [54:55] little issue in my driver node. Looks [54:59] like it crashed last time we ran it. [55:02] Yeah, don't do live demos. They never [55:03] work. [55:05] >> [laughter] [55:05] >> Cool, we'll move on and we'll get back [55:07] to it if we have time. But I don't want [55:08] to waste our time debugging stuff. But [55:10] essentially [55:11] um [55:13] uh [55:15] Yeah, so this is all the launch file [55:17] that we could run for this demo. [55:24] So it would look something like this, [55:25] right? So this is the four nodes that [55:27] you're going to run. One of them is [55:28] actually using like a another launch [55:30] file in another package. And this is [55:32] like the three other nodes. So it's like [55:33] just you know the more exciting more [55:35] complicated version of what you've seen [55:37] before. [55:39] >> Wait. [55:40] >> Yeah. [55:40] >> Can you go back? [55:42] Can you say that again? Like what's the [55:43] new file? [55:45] >> Sure. Okay. [55:46] Um so [55:48] there's two ways, right, of writing [55:49] these launch files. So, you can either [55:51] include every node that you have, but [55:53] then you can think of like there's [55:54] sometimes where uh [55:56] someone has actually already like [55:58] written a launch file for a subsystem of [56:00] your of your of your robotic stack. So, [56:04] that subsystem can be launched with [56:06] itself a launch file. And so, you can [56:08] now have like you can build a hierarchy [56:10] of launch files that uh each call each [56:12] other. It's unlikely that you're going [56:14] to do or maybe you will. Um [56:16] >> using like open source code? [56:20] >> You can. You can. Um so, I wasn't in [56:22] this specific like what I was trying to [56:23] run, uh but you certainly can. And uh [56:28] but so but uh it's kind of up to you. [56:30] Like the launch file their launch file [56:31] might uh define some parameters, right? [56:33] And we're actually overriding some of [56:35] the parameters of their launch file. So, [56:36] we're assuming like video devices, [56:38] whatever argument we have defined at the [56:39] top. So, it depends what you want to do. [56:42] Uh and then yeah, sometimes it's just [56:43] easier Some of these launch files are [56:44] huge. Like some of them define like, you [56:47] know, like the image width of every [56:48] single like 16 cameras that you have. Um [56:51] so, it's kind of nice to just be able to [56:52] run someone else's launch file who went [56:54] through like the [56:55] pain painstaking [56:57] uh like the really painful work of like [56:59] writing it. Um and then but but uh [57:01] sometimes it just makes sense to write [57:02] everything from scratch from like the [57:03] node. [57:05] Mark that sure. [57:07] Uh-huh. [57:08] >> Uh earlier you mentioned that some [57:09] parameters are stored for us in master [57:11] and you get access to those. Are the [57:12] parameters that we're changing here [57:13] those same parameters or are they [57:14] something else? [57:16] >> Are they the same parameters? Sorry, can [57:18] you repeat that question one more time? [57:19] >> Uh you mentioned that like some things [57:21] like the rate for example that you have [57:22] in master and then you have access to [57:24] those. Are these the same parameters [57:25] that they're the ones that are in master [57:27] like this team needs change at the [57:29] global level? [57:30] >> Yeah, it it's a good question. Yeah, it [57:32] depends how you define them. They can be [57:34] I mean they're like they're if you [57:35] define any parameters here, some of them [57:37] are [57:39] um [57:39] Yeah, it's a little subtle, but like [57:41] there some of the parameters that you [57:42] define inside like nodes, for example, [57:46] will get like a prefix to them. [57:49] Um they will still be accessible to [57:51] everyone, but they're just kind of like [57:52] an indicator that like this is for this [57:55] node, so you might end up with a [57:56] parameter that looks something like [57:57] image view two {slash} [58:00] uh [58:01] {slash} like auto size or something. Um [58:04] so it's a little there there are Yes, [58:05] all the parameters like live at the same [58:07] place in in the master, um but there's [58:10] like a soft enforcement of some [58:12] parameters living in one node or [58:14] another. [58:16] Cool. Okay. This is a launch file. [58:19] Um [58:20] Okay, developing with ROS. [58:22] So ROS has a lot of development tools. [58:24] Uh one thing that you will deal with is [58:26] the catkin workspace. [58:27] Uh [58:28] what the catkin workspace means is uh so [58:31] this seems more complicated than it [58:33] feels like you need like it feels like [58:34] you should be able to just be able to [58:35] write your Python scripts and you should [58:37] just be able to run them. It's kind of [58:38] true, [58:39] uh but keep in mind that ROS is also [58:41] built to to support all these different [58:43] languages, uh including C++ and C++ [58:46] needs to be compiled. And so they had to [58:48] kind of build this Well, they didn't [58:49] have to, but uh they decided to build [58:51] this awesome tool which is called your [58:53] catkin workspace. Uh the the the catkin, [58:55] which is basically a wrapper for [58:57] compiler, uh and then make sure and it [58:59] just like organizes all your packages [59:01] for you. Uh but you can think of like [59:02] catkin workspace as just a nice way to [59:04] organize your project so that compilers [59:06] can find the different parts that you [59:07] need. Uh [59:08] and yeah, I mean those are like the [59:09] commands that's going to look like. [59:10] Basically like you'll create a directory [59:11] Your your workspace is just a directory [59:13] in your on your laptop or whatever. [59:16] Uh and then you know, you'll go in your [59:18] workspace and you can [59:19] uh compile it. Uh compile like any of [59:21] the the programs that are there. [59:25] Cool. Um [59:27] okay, the programs that are there are [59:28] going to be in your workspace are going [59:30] to be organized in packages. So, if you [59:33] remember like this restaurant command, [59:35] we had like restaurant package name and [59:37] then node. So, you might have like [59:39] restaurant like 8274 is maybe your [59:42] package name and then you'll have like [59:44] talker as like your node. [59:46] Uh, and so like the package are [59:49] uh, going to be structured in a specific [59:50] way, which again feels overkill, but [59:53] it's, you know, blame C++ for that. Uh, [59:57] uh, but like you know, it's going to [59:58] look something like this. So, you're [59:59] going to have a a big folder where [60:01] you'll have these files that you don't [60:03] have to care about. They're like C make [60:04] files. You're going to have usually a [60:05] con- configuration folder that has a [60:07] bunch of configs, uh, include files your [60:09] header files in C++. Your scripts, [60:11] that's probably the only one you care [60:12] about. This little have like your Python [60:14] scripts that are your nodes, [60:16] essentially. So, you'll have your catkin [60:18] workspace. Inside your catkin workspace, [60:20] you'll have a source folder. Inside the [60:21] source folder, you'll have your [60:23] packages. [60:24] And like in one of your package, you'll [60:26] have a scripts folder. [60:29] And then in that folder, you'll have [60:31] finally your little Python script that [60:33] is doing something. [60:34] Uh, and that's true like now if you have [60:35] like C++ source code, it's going to be [60:37] like in the source folder. You'll have [60:38] your launch file, which can be in your [60:40] own launch folder. Uh, you'll have [60:42] messages, so if you're defining new [60:43] types of messages, uh, uh, then you can [60:46] do that. And then, um, [60:49] yeah, that one we're not going to talk [60:50] about cuz it complicates a little bit. [60:52] Uh, [60:52] and yeah, [clears throat] you can, you [60:53] know, ROS comes with a bunch of stuff. [60:55] Uh, and for example, like comes with a [60:56] nice little utility to create packages [60:59] and tell you. So, what this does for [61:00] example, it tells you like, "This is the [61:02] name of what I'm trying to the package I [61:04] want to create and these are the [61:05] dependencies that it has." Uh, [61:08] yeah. [61:10] >> Is that auto in C++ or is that just [61:12] where we would put the C++ files if we [61:14] coded in C++? [61:15] >> So, if you were to code in C++, you [61:17] would put all your source in like the [61:19] source folder. [61:20] >> [snorts] [61:20] >> And then, uh, [61:22] catkin make it you'd don't have to like [61:24] go out of that directory and then you [61:26] run like catkin make and it'll find the [61:28] project you have and it'll compile it. [61:29] >> But if we write in Python, it'll run [61:31] Python. [61:32] >> Right. Yeah. There's no like conversion. [61:36] Yeah. [61:37] And something that will happen to you, [61:40] uh [61:41] I promise, we've thought this for a [61:42] couple years now. You will run like [61:45] rosrun package name like A274 turtlebot [61:48] and then you will uh try to run your [61:51] script name which is just like you'll [61:53] you'll just add it a file there [61:54] talker.py and it won't run. [61:56] Uh and you won't know why. It won't find [61:58] it. It'll be like, "Hey, this doesn't [61:58] exist." You have to go in there and make [62:00] it executable which uh you know, all the [62:03] homework is going to walk you through [62:04] how to do that but just keep that in [62:06] mind. When it doesn't find it, make sure [62:08] that your script is executable. Uh it's [62:10] just like one command that you'll run [62:11] with like that that file. Uh but [62:13] remember this, it's going to come up. [62:15] Yeah. [62:17] >> Do you use it make functionality of node [62:19] or a Python [62:21] >> Yeah, good question. Yeah, no that's a [62:23] great question. Um [62:25] Like I mean so I think that's kind of [62:27] like just a software engineering kind of [62:29] decision on your end. Uh [62:31] you know, the packages usually are at [62:34] kind of the capability level. Uh people [62:37] use capability in robotics a lot. I [62:39] don't think I've ever heard it defined [62:41] but I feel like we kind of all know what [62:42] it means. Um so like you might have like [62:44] an estimation package. Uh and that [62:46] estimation package is you know, an [62:47] abstraction of all your your estimation [62:49] capability and then inside of that [62:52] they're going you're going to have [62:52] different nodes. Um but it's kind of up [62:54] to you as to like what you think is a [62:56] package. Usually it should just be like [62:58] a pretty reusable, you know, set of [62:59] tools that that like address like a [63:01] specific need for for your robot. But [63:05] but yeah, that's that's a little bit [63:06] like project dependent. Hopefully that [63:08] answers your question a little bit. [63:09] Yeah. [63:09] >> Wait, so the whole blue thing I can [63:11] compare this package like [63:13] a folder to this. [63:15] >> Yeah, I know that would be a package. [63:16] Yeah. Yeah, the blue thing would be your [63:18] package. [63:19] >> So like [63:22] Yes. So, you have your workspace that [63:25] usually has a source folder and inside [63:26] that you'll have multiple blue things [63:29] like this. [63:32] Uh they're not You don't You don't [63:34] necessarily need them, right? Like you [63:35] could have a bunch of launch files. [63:37] Um [63:40] Yeah, you could have also each package [63:42] like the So, the launch files are in [63:43] packages. Um but something that's cool [63:46] is that the launch file also knows how [63:48] to find things in other packages. So, [63:50] you could think of having [63:52] uh you could hang out having like a [63:54] package that is just like this is like [63:55] sensor driver package. And it just has a [63:57] bunch of nodes that are just like [63:58] sensor drivers. And then you have [64:00] another package that's like your [64:01] estimation package. You have a launch [64:02] file inside your estimation package that [64:05] knows that like oh, if I want to run an [64:07] UKF that is based off my lidar, like [64:10] when I [64:11] write this, right? It'll say find A274. [64:15] It could be like find sensor driver [64:17] package instead of A274 and then say [64:20] like in that package there's a launch [64:21] file that launched the that launches the [64:23] lidar. So, I want you to go launch that [64:25] and then [64:26] uh and then launch the rest of the [64:27] stuff. [64:28] Yeah. [64:29] Okay. [64:31] Um yeah, we're just going to go through [64:33] a couple of like tools I guess with the [64:34] couple of minutes we have left uh that [64:36] you might come up mostly as a reference. [64:39] Uh this is like a lot of information. [64:42] I [laughter] don't know like I don't [64:43] think anyone expects you to retain all [64:45] of this. Uh but it's just like hopefully [64:47] this will be in the back of your mind [64:49] when it comes up. Uh yeah, so debugging [64:51] there's all the things that ROS does. [64:53] Oh, it's hard. Um it's harder than [64:56] debugging your you know, self-contained [64:59] Python script. Uh but there's a lot of [65:01] tools to help you do that. One of them [65:03] is this console. Uh so, when we were [65:05] doing the logging command, right? As I [65:07] was saying it's kind of like a print [65:09] uh statement, but it's a little more [65:10] sophisticated than that. You can log it. [65:12] You can also and look at them on this [65:15] nice little GUI that's telling you [65:16] whether something is very important or [65:18] not important. [65:20] Uh and you know, that's like the type of [65:21] stuff that you know, these like 12 [65:22] people are in the room with the 12 [65:24] screens that we had earlier. They are [65:26] probably looking at at some of these [65:28] these tools to like make sure that like, [65:29] oh, my lidar or my USB camera is not [65:32] working now anymore for some reason. [65:34] Um so that's like normal the type of [65:35] stuff that you'll have. You also want to [65:37] learn to use the debugger a lot. Um that [65:40] might be something you've used before. [65:41] So that just gives you a way of like [65:43] stepping through your code line by line [65:45] in the code. Uh PDB is awesome. [65:47] Definitely learn to use it. Uh [65:50] the other thing, write your own [65:51] messages. There's a lot of of [65:54] uh basic messages already exist in ROS. [65:56] I don't think that I remember the class [65:58] requiring you to write any custom one, [66:00] but it might. [66:01] Um [66:02] So like I said, uh these these uh so ROS [66:05] can support all these different [66:06] languages, right? C++, Python, uh Java, [66:10] like Julia. Uh and so but you need like [66:13] all of that needs to be able to send [66:14] these messages, right? So you have to [66:15] have a way of now defining what a [66:17] message is that is uh language agnostic. [66:20] Essentially that doesn't depend on which [66:21] language you're defining it in. So it [66:23] cannot be just like a message cannot [66:24] just be a Python class. So you have this [66:27] other syntax that allows you to define [66:29] objects, and then ROS comes with tools [66:31] that, you know, if you define a message, [66:33] let's say let's say we'll have a student [66:34] message, then we'll have like a first [66:36] name, a last name, age, grade. Uh [66:38] and then you'll have these tools that [66:40] will allow you to take this and then [66:42] automatically generate the Python class [66:43] that you need to send messages with it, [66:45] automatically generate the C++ class [66:47] that you need to send messages or or the [66:49] Java [66:51] um [66:51] class or the Julia everything. So so [66:54] that's what that's like one of the [66:55] strength of of using this pub/sub thing [66:57] too is that like it becomes like very [66:58] language agnostic as long as you have [67:00] these tools that that come with ROS that [67:01] allow you to [67:02] create use a syntax, create messages, [67:04] and then just send them over any [67:06] language. Um [67:08] Yeah, okay. This is going to like muddy [67:11] things a little bit. I said Ross is like [67:13] just pub/sub. Some people like to use [67:15] different things in pub/sub for specific [67:17] application. Ross can also support that. [67:20] Uh I would argue that like 95% of Ross [67:22] usage is through pub/sub. So, I think [67:25] it's still fair to say that Ross is [67:26] basically pub/sub and that's what it was [67:27] originally. But just keep in mind that [67:29] like if you're thinking of it as a [67:30] different design pattern [67:32] uh than pub/sub, like Ross might support [67:34] it. Just look into it uh and that might [67:36] be something you want to you want to [67:37] use. [67:38] Um parameter server, I think we talked [67:40] about it already. [67:42] Uh [67:43] Okay, this is something you will see [67:45] also. I think probably under appreciated [67:49] you know, feet of of uh I don't know, [67:52] like tools or modeling, whatever you [67:53] want to call it. Uh [67:55] So, one way to describe your robot that [67:57] is super powerful is as as these using [68:01] these XML [68:02] uh file that are called URDF files. So, [68:05] turns out it's actually if you like [68:06] think hard hard enough about it, it's [68:08] actually possible for almost all robots [68:11] with some exception obviously uh to [68:14] write like all your robots as as [68:15] something called like a kinematic chain. [68:17] So, maybe some of you took uh Professor [68:18] Khatib's class, which I think talks [68:20] about this. Uh but like you'll basically [68:22] think of like, you know, you will have a [68:23] bunch of of uh links that have joints in [68:26] between them. So, basically just think [68:27] of like a joint is like uh you know, [68:30] your elbow and then the links is like [68:31] your forearm and your upper arm. Uh and [68:34] so, like you have you can think of how [68:36] and then it becomes [clears throat] kind [68:37] of this tree structure, right? So, a [68:38] robot like Atlas can actually be like [68:40] described in just this pretty abstract [68:42] way. Uh and turns out if you take this [68:44] description and you put that in XML [68:46] file, then that becomes like a super [68:48] reusable way of modeling your robot. Um [68:51] and that's super cool because now if [68:54] someone is working on a different robot, [68:56] so someone has like [68:58] you know, they have like a Jackal robot, [69:00] Jackal robot or whatever the Stanford [69:02] robot is called. And maybe you have like [69:04] software for it and you're running it in [69:06] simulation and you have this description [69:07] of the file. And then someone else comes [69:09] around and is just like, "Ah, I wonder [69:10] how my robot, you know, I have Atlas, [69:12] it's like way bigger, maybe cooler." Uh [69:16] and then you kind of want to know, you [69:17] know, does the algorithm work well with [69:19] on their robot? [69:21] Uh there's a lot of things obviously [69:22] that can go into it, but basically to [69:23] get like their robot in simulation or [69:25] even to get an understanding of their [69:27] robot could literally just be to get [69:29] their URDF. So, to get their URDF is [69:32] this XML file. Uh that's what we call [69:33] them, Universal Robot Description [69:35] Format. So, this XML file that describes [69:37] the robot. Uh and then you can like put [69:39] that in your stack and then you have [69:41] your simulation, uh you know, [69:42] description of kinematics, dynamics, all [69:44] this stuff can come out of just this [69:46] file. So, look for cool URDF online. [69:48] There's plenty of them. Uh people like [69:50] come up with with weird ones all the [69:51] time. Uh and that gives you like a way [69:53] of modeling robots. [69:55] Uh you will interact with Gazebo. So, [69:57] those URDF file, right? Uh [70:00] like describe robots. Like for example, [70:02] this little [70:03] I forgot how this guy is called, but you [70:04] know, this little robot here. Um [70:07] they can [clears throat] This is the PR2 [70:08] actually that I was talking about [70:09] earlier from Willow Garage. Uh so, all [70:11] these URDF describe robots and you can [70:12] load them in a simulator that's called [70:14] Gazebo. Gazebo is well integrated with [70:16] ROS. Uh it's not ROS. They're kind of [70:19] annoying about this, but but uh they're [70:21] like very well integrated and basically [70:23] everyone like puts them together. So, uh [70:25] regardless of them liking it or not, [70:27] they're kind of part of the ROS [70:29] ecosystem. And you will use it and it's [70:31] very a very powerful way of modeling [70:33] your your system. The interesting thing, [70:36] the thing that is [70:37] uh there's two things I think that like [70:39] matters a lot. So, [70:41] one is that yes, you will get something [70:43] that, you know, you will simulate the [70:44] physics of your system. That's [70:46] important. You're going to make sure [70:47] that like your controller is working, [70:48] that your path finding uh your path [70:50] following controller is working. The [70:52] other thing that's important of a [70:53] simulator, like Gazebo, have a good [70:55] simulator is not just that it simulates [70:57] physics properly, but it's also that [70:59] it's representative of the software [71:01] environment that you will run your code [71:03] in. So, and the reason why that matters [71:06] is that because [71:07] if you you'll you'll spend a bunch of [71:10] time writing your code so that it works [71:11] in simulation, and then if basically it [71:13] works in simulation, but now to make it [71:15] work on the real robot you have to [71:17] rewrite everything from scratch, [71:19] um [71:19] I you know, the simulation really gave [71:21] you very little. But if it the [71:24] simulation is structured [71:25] like Gazebo is so that you can run your [71:28] ROS stack against the simulation, and [71:31] then literally take the same code that's [71:33] publishing the same messages the same [71:35] node, and then just put that on the real [71:36] robot, and then [71:38] and the real robot knows how to you [71:39] know, communicate with the same stack [71:41] without changing anything, then you just [71:42] save yourself like weeks of development, [71:44] right? Um so, that's super cool, and [71:47] you'll see that is true for the [71:48] Trollbot. Basically, [71:50] uh you'll have to do a lot of uh [71:52] hardware, you know, development at the [71:54] end of the project, but use Gazebo a [71:56] lot. Like you you can work from home [71:59] basically. I mean, you should come to [72:00] the lab. I'm I'm not going to say people [72:02] have to [72:02] >> [laughter] [72:02] >> come to the lab. But like you can save [72:04] so much time by just making sure it [72:06] works in simulation first. If it doesn't [72:07] work in simulation, [72:09] it's probably not going to work on the [72:10] real robot. And then if it works in [72:12] simulation, it might work on the real [72:14] robot. [72:15] Um so, yeah, keep that in mind. Use [72:17] simulation, it's super powerful, it'll [72:18] save you a bunch of time. [72:20] Uh there's other stuff you'll hear about [72:22] the transform library and stuff. Uh all [72:25] these things [72:26] go [72:27] find information [72:29] uh online. There's uh ROS has a wiki. [72:31] GitHub is great. The readme pages of all [72:33] these packages. Stack Overflow. Just [72:35] Google things. Talk to TAs, they all [72:38] know ROS very well. Um yeah, tons of [72:40] resources. Quick note on ROS 2. [72:43] So, ROS has been around since 2007. [72:46] Uh that's more than a decade. It's like [72:48] so much time in software [72:51] uh land. So, ROS has actually done a [72:53] rehaul like a overhaul of of everything [72:56] recently in something called Ross 2. It [72:58] doesn't support everything yet. So, the [73:00] class is, you know, super honest like [73:02] very careful at using cutting edge [73:04] things, but he also doesn't want to [73:06] you know, make you guys use something [73:07] that isn't super mature. So, uh we're [73:10] going to stay away from Ross 2 for now, [73:11] but you know, [73:12] >> [snorts] [73:12] >> uh [73:13] uh keep an eye out for that for it. It's [73:16] it has a lot of cool things that are [73:17] coming out. [73:18] Um yeah, so just wrapping up uh real [73:20] quick. Uh PSet 1 released Friday. [73:23] Uh yeah, you'll have through next week [73:26] you're going to work through basic [73:27] simulation and the setup of all of this. [73:29] Um and yeah, sign up for the seminar [73:31] class as well, which is also Friday. If [73:33] this is this class isn't going to teach [73:35] you how to build robots. Now, if you [73:36] want to get ideas on what research to do [73:40] uh with all these skills, the Stanford [73:42] Robotics Seminar runs every Friday at [73:44] 11:00 a.m. And it's super casual. You [73:47] get food. [73:49] Um you get one credit to attend. And uh [73:52] you get really good talks and research [73:55] ideas and potential, you know, future [73:57] advisors, who knows. Um so yeah, that's [73:59] it.