Boston Dynamics Atlas Robot Hits 80% Success Rate in Gymnastics

Boston Dynamics

Pairing trajectory optimization with model predictive control slashes routine development time while sustaining an 80% execution rate across whole-body maneuvers.

AI-assisted summary of More Parkour Atlas — original source. Check the original for full context.

Source published: 2019-09-24

Key ideas

Boston Dynamics demonstrated its humanoid robot Atlas performing a continuous gymnastic parkour routine utilizing full whole-body coordination. (0:00)
The development workflow employs an optimization algorithm to convert high-level maneuver descriptions into dynamically feasible reference motions. (0:00)
A real-time model predictive controller tracks reference trajectories while smoothly blending transitions between sequential maneuvers. (0:00)
The new streamlined development pipeline produced dynamic routines significantly faster than prior methods, achieving an operational success rate of roughly 80%. (0:00)

Detailed notes

## Background Published on YouTube by Boston Dynamics on September 24, 2019, this demonstration features the humanoid robot Atlas performing an athletic gymnastic routine. The video description details a streamlined development process where an optimization algorithm and a model predictive controller allow Atlas to coordinate its whole body across dynamic maneuvers with an 80% success rate.

## TL;DR The video highlights the humanoid robot Atlas executing a continuous series of gymnastic and parkour maneuvers using coordinated movements of its arms, legs, and torso to sustain momentum. To achieve this, an optimization algorithm converts high-level maneuver descriptions into dynamically feasible reference trajectories. In execution, a real-time model predictive controller smoothly blends these trajectories from one maneuver directly into the next. This refined engineering approach significantly reduced routine development times while achieving an approximate 80% operational success rate.

## Atmosphere & style The presentation has a sleek, clinical, and astonishingly fluid feel, emphasizing pure robotic agility without distracting production elements. Watching Atlas carry out complex athletic choreography creates a striking sensation of mechanical life and physical grace. The brief showcase offers an intense, focused glimpse into advanced robotics that leaves a lasting impression of effortless kinetic power.

## Bulletin - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) Boston Dynamics demonstrated its humanoid robot Atlas performing a continuous gymnastic parkour routine utilizing full whole-body coordination. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) The development workflow employs an optimization algorithm to convert high-level maneuver descriptions into dynamically feasible reference motions. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) A real-time model predictive controller tracks reference trajectories while smoothly blending transitions between sequential maneuvers. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) The new streamlined development pipeline produced dynamic routines significantly faster than prior methods, achieving an operational success rate of roughly 80%.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) Atlas executes a continuous sequence of dynamic gymnastic maneuvers rather than isolated, discrete actions. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) The robot relies on whole-body coordination—synchronizing its legs, arms, and torso—to preserve momentum and dynamic balance. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) New development techniques were implemented to streamline the creation of complex robotic parkour routines. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) High-level behavioral descriptions of movements are computationally converted into dynamically feasible reference trajectories using an optimization algorithm. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) Real-time trajectory tracking on Atlas is managed by a model predictive controller. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) The model predictive controller enables seamless blending from one gymnastic maneuver directly into the next. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) The optimization and control pipeline drastically reduced routine development time compared to earlier Atlas development cycles. - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) The dynamic routine achieves a performance success rate of approximately 80% under this control framework.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) Dynamic Gymnastic Maneuvers and Motion Control

During the brief clip, a prompt notes, “I'm going to go to the next video,” as the showcase captures Boston Dynamics' humanoid robot, Atlas, executing a continuous sequence of dynamic gymnastic maneuvers. Rather than relying solely on isolated leg movements, Atlas employs whole-body coordination—synchronizing its legs, arms, and torso—to maintain balance and momentum across complex parkour actions.

The development workflow behind these maneuvers relies on a streamlined two-stage process. First, an optimization algorithm converts high-level behavioral descriptions of each movement into dynamically feasible reference trajectories. Next, Atlas executes and tracks these paths in real time using a model predictive controller, which blends individual actions seamlessly into the next movement. This streamlined pipeline allowed the team to construct dynamic routines significantly faster than during earlier Atlas development cycles, yielding an operational success rate of approximately 80%.

```mermaid flowchart LR accTitle: Atlas Dynamic Motion Control Pipeline accDescr: Diagram illustrating the workflow from high-level maneuver descriptions to dynamic routine execution on Atlas. subgraph Planning["Trajectory Generation"] A["High-Level Maneuver Description"] --> B["Optimization Algorithm"] B --> C["Dynamically Feasible Reference Motions"] end subgraph Execution["Real-Time Control"] C --> D["Model Predictive Controller"] D --> E["Whole-Body Dynamic Routine"] end ```

- Atlas utilizes its entire physical structure—including torso, arms, and legs—to coordinate momentum and balance throughout the routine. - Dynamic reference trajectories are generated by an optimization algorithm from high-level movement inputs. - Real-time tracking is handled by a model predictive controller that smoothly blends transitions between distinct maneuvers. - The development approach achieved an approximate 80% success rate while drastically reducing iteration time compared to prior routines.

## Mentioned Atlas, Boston Dynamics

## Video description Atlas uses its whole body -- legs, arms, torso -- to perform a sequence of dynamic maneuvers that form a gymnastic routine. We created the maneuvers using new techniques that streamline the development process. First, an optimization algorithm transforms high-level descriptions of each maneuver into dynamically-feasible reference motions. Then Atlas tracks the motions using a model predictive controller that smoothly blends from one maneuver to the next. Using this approach, we developed the routine significantly faster than previous Atlas routines, with a performance success rate of about 80%. For more information visit us at https://www.BostonDynamics.com.

[Source](https://www.youtube.com/watch?v=_sBBaNYex3E)

中文简报

波士顿动力Atlas机器人体操动作成功率达80% 轨迹优化与模型预测控制的结合大幅缩短了动作编排开发时间，同时在全身机动动作中保持了80%的执行成功率。该视频展示了人形机器人Atlas如何通过手臂、腿部和躯干的协同运动来保持动量，连续完成一系列体操和跑酷动作。为实现这一目标，一种优化算法将高层级动作描述转化为动力学可行的参考轨迹。在执行阶段，实时模型预测控制器将这些轨迹平滑衔接，直接从一个动作过渡至下一个动作。这种经过改进的工程方法显著缩短了动作套路的开发时间，同时实现了约80%的运行成功率。波士顿动力展示了其人形机器人Atlas利用全身协调能力执行连续的体操跑酷套路。 (0:00) 开发流程采用优化算法将高层级动作描述转换为动力学可行的参考运动。 (0:00) 实时模型预测控制器在追踪参考轨迹的同时，平滑衔接连续动作之间的过渡。 (0:00) 全新精简的开发流程生成动态套路的速度显著快于以往方法，并达到了约80%的运行成功率。 (0:00)

Reading summary

Boston Dynamics demonstrates that its humanoid robot, Atlas, can execute continuous gymnastic routines through whole-body coordination supported by a two-stage motion control pipeline. Whole-Body Coordination in Dynamic Maneuvers (0:00) Atlas executes continuous parkour and gymnastic sequences by coordinating its torso, arms, and legs simultaneously rather than relying solely on isolated leg movements. This unified physical synchronization allows the robot to actively manage momentum and maintain balance across complex transitions. Trajectory Optimization and Real-Time Control (0:00) The motion development workflow employs an optimization algorithm that translates high-level movement descriptions into dynamically feasible reference paths. A real-time model predictive controller then tracks these trajectories, smoothly blending distinct movements into a continuous routine. This streamlined approach significantly accelerated routine construction compared to prior Atlas development cycles and achieved an operational success rate of approximately 80%. Although the streamlined pipeline drastically reduces iteration time and reaches an approximate 80% success rate, the performance margin highlights ongoing challenges in achieving complete operational reliability during dynamic routines.

精炼阅读摘要

波士顿动力（Boston Dynamics）展示了人形机器人Atlas执行连续动态体操与跑酷动作的能力，其核心在于结合全身协同运动与“离线轨迹优化结合实时模型预测控制”的两阶段控制流程，显著提升了动作编排效率并将运行成功率提升至约80%。全身协调与动态动量控制 (0:00) 在跑酷动作的执行中，Atlas并非仅依靠局部的腿部动作，而是全面动用躯干、手臂和双腿等整体物理结构。通过全身各部位的同步协调，Atlas能够在复杂的翻滚、跳跃等体操动作中持续调控动量并维持机体平衡，确保了动作的连贯性与物理稳定性。双阶段优化管线与实时控制 (0:00) Atlas动作开发的底层依托于一套高度精简的两阶段工作流：首先，系统通过优化算法将高层级的人类行为描述转化为在动力学上切实可行的参考运动轨迹；随后，机器人利用模型预测控制器（MPC）进行实时轨迹跟踪，并将独立的单一动作无缝衔接至下一个动作。波士顿动力表示，相比早期的Atlas开发周期，这一新型控制管线大幅缩短了动态动作序列的开发与迭代时间，使整套复杂动态跑酷动作的实际运行成功率达到了约80%。该流程验证了离线轨迹生成结合实时模型预测控制在人形机器人复杂动作中的实用性与开发效率，但约80%的成功率也反映出系统在面对极端动态运动时仍面临一定的稳定性与可靠性边界。

Detailed notes — zh

## Background 该视频由波士顿动力（Boston Dynamics）于 2019 年 9 月 24 日发布在 YouTube 平台上。视频简介指出，双足机器人 Atlas 调动全身（腿部、手臂、躯干）完成了一套连续的体操动作，团队通过优化算法与模型预测控制器（MPC）等新技术精简了开发流程，使得动作开发显著提速并达到了约 80% 的执行成功率。 ## TL;DR 视频展示了 Atlas 机器人呈现的高难度全身跑酷与体操连续动作，包括流畅串联的翻滚、跳跃与旋转等机动。研发团队通过优化算法将高层动作描述转化为动力学可行的参考轨迹，并利用模型预测控制器（MPC）实现动作之间的平滑融合与实时姿态跟踪。这套全新的流线型开发流程大幅提升了机器人动作编排的效率，在实际演练中保持了约 80% 的动作成功率。 ## Atmosphere & style 整段视频简短而极具冲击力，画面直观呈现了 Atlas 行云流水般的动态动作，展现出前沿机器人的敏捷与力量感。背景中仅夹杂着简短的工作人员画外音，没有任何多余的修饰或解说，烘托出一种专注于工程成果展示的严谨与震撼氛围。观看过程让读者直观感受到拟人化机械在物理空间中展现出的惊人协调性与突破性进展。 ## Bulletin - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 波士顿动力展示了双足人形机器人 Atlas 凭借腿部、手臂与躯干的全身协同，完成了一套包含翻滚、跳跃与旋转的流畅动态体操套路。 - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 研发团队引入了优化算法与模型预测控制器（MPC）相结合的技术流程，将高层动作描述转化为动力学可行的参考轨迹并实现动作间的平滑过渡。 - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 得益于流线型的全新开发流程，Atlas 动作序列的研发速度较以往显著提升，整套体操动作的执行成功率达到约 80%。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 波士顿动力通过视频展示了 Atlas 机器人的高难度跑酷与体操动作序列。 - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) Atlas 充分调动了手臂、腿部与整个躯干的协同能力，以维持高动态运动下的物理平衡。 - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 整套操练涵盖了连续的动态机动动作，包括翻滚、跳跃以及全身旋转。 - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 团队采用了全新的流线型开发技术，显著缩短了复杂机器人动作套路的研发周期。 - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 规划阶段利用优化算法，自动将高层级的动作意图描述转换为动力学可行的参考轨迹。 - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 控制层引入模型预测控制器（MPC），实时跟踪生成的参考运动轨迹并实现平稳调整。 - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 模型预测控制器（MPC）能够在不同动作机动之间进行平滑融合，避免动作衔接时的突变与失稳。 - [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) 依靠该控制与规划架构，Atlas 在执行这套复杂的体操套路时达到了约 80% 的实操成功率。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=_sBBaNYex3E&t=0s) Atlas 机器人全身动态跑酷与体操动作展示在本段简短的视频实录与演示中，伴随画外音提示“I'm going to go to the next video. you”，波士顿动力（Boston Dynamics）展示了双足人形机器人 Atlas 展开的一连串高难度全身跑酷与体操动作规范。Atlas 充分调动了其腿部、手臂以及整个躯干的协同运动能力，将翻滚、跳跃与旋转等一系列动态机动动作无缝串联，形成了一套流畅连贯的动态体操套路。与以往的动作编排方式不同，研发团队采用了全新的流线型开发技术，大幅缩短了机器人动态行为的研发周期。整个控制架构首先通过优化算法将对每个动作的高层级描述转化为动力学可行的参考运动轨迹；随后，Atlas 借助模型预测控制器（Model Predictive Controller, MPC）对这些参考轨迹进行实时跟踪，确保各动作动作之间能够平滑自然地过渡衔接。依托这一技术流程，Atlas 动作序列的开发速度相较以往显著提升，并在实际演练中达到了约 80% 的动作成功率。 ```mermaid flowchart TD accTitle: Atlas 跑酷动作生成与控制流程 accDescr: 展示从高层动作描述到最终全身动态执行的控制流架构 subgraph Planning["规划阶段"] A["高层动作描述"] -->|"优化算法"| B["动力学可行参考轨迹"] end subgraph Execution["控制与执行"] B -->|"轨迹跟踪与融合"| C["模型预测控制器 (MPC)"] C -->|"全身协同控制"| D["Atlas 硬件执行 (躯干/四肢)"] end ``` - **全身协同运动**：Atlas 在执行体操动作时综合运用手臂、腿部和躯干，保持高速动态平衡。 - **轨迹优化机制**：采用专门的优化算法，将抽象的高层动作指令转化为符合物理动力学特性的参考轨迹。 - **平滑过渡控制**：依托模型预测控制器（MPC）实现机动动作之间的平稳混合与实时调整。 - **开发效率与成功率**：全新流程使动作开发周期大幅缩短，整体动作套路的执行成功率约为 80%。 ## Mentioned Atlas, 波士顿动力 (Boston Dynamics), 模型预测控制器 (MPC) ## Video description Atlas uses its whole body -- legs, arms, torso -- to perform a sequence of dynamic maneuvers that form a gymnastic routine. We created the maneuvers using new techniques that streamline the development process. First, an optimization algorithm transforms high-level descriptions of each maneuver into dynamically-feasible reference motions. Then Atlas tracks the motions using a model predictive controller that smoothly blends from one maneuver to the next. Using this approach, we developed the routine significantly faster than previous Atlas routines, with a performance success rate of about 80%. For more information visit us at https://www.BostonDynamics.com. [Source](https://www.youtube.com/watch?v=_sBBaNYex3E)

Original transcript

[0:00] I'm going to go to the next video. you