波士顿动力公司，这家著名的人形机器人研发机构，近日揭秘了其最新人形机器人Atlas的研发历程。Atlas能够完成高难度动作，如跑酷、后空翻等，这背后是工程师们六年来的经验与教训总结。
Atlas的最新版本采用了电动驱动，身材更加小巧灵活，这标志着波士顿动力在适应工业环境方面的研发方向转变。该公司计划在明年初在韩国现代汽车工厂进行试点测试，并预期几年后实现全面投产。
波士顿动力机器人工程师Robin Deits在机器人顶会RSS的技术分享会上，详细介绍了Atlas的研发历程和从中学习到的经验教训。Deits是MIT博士，自2018年以来一直在波士顿动力工作，专注于Atlas人形机器人的控制。
在分享中，Deits重点介绍了MPC（模型预测控制）技术在Atlas中的应用。MPC是一种高级控制策略，能够处理多变量系统并考虑约束条件，是Atlas实现各种动作的核心技术。Deits表示，自2019年以来，所有Atlas的动作都是通过MPC实现的，包括跑酷、体操、跳舞、后空翻等。
MPC版本的Atlas有几个共同特点：包括非线性动力学的融入、迭代线性化和求解QP、不追求收敛、利用问题结构提高速度、不将求解器视为黑盒等。Deits和他的团队已经在机器人上解决了大量QP问题，并在模拟环境中解决了更多问题，这些努力确保了Atlas能够在复杂环境中提供鲁棒的控制解决方案。
尽管Atlas在技术上取得了巨大进步，但机器人仍然可能会翻车。工程师们从这些失败中吸取教训，不断优化控制策略，以提高机器人的稳定性和可靠性。随着技术的不断进步，Atlas和其他人形机器人的未来应用前景令人期待。
英语如下：
News Title: “Boston Dynamics Unveils: The Six-Year Challenge and Achievement Behind Atlas’s Difficult Maneuvers”
Keywords: Boston Dynamics, Humanoid Robots, Difficult Maneuvers
News Content:
Boston Dynamics, a renowned institution for the development of humanoid robots, recently unveiled the development journey of its latest humanoid robot, Atlas. Atlas is capable of performing difficult maneuvers such as parkour, backflips, and more. Behind these feats are the experiences and lessons learned by engineers over six years of work.
The latest version of Atlas employs electric drive and is designed to be more compact and agile, marking a shift in Boston Dynamics’ research direction towards adapting to industrial environments. The company plans to conduct pilot tests in the Hyundai Motor Group’s factory in Korea early next year and anticipates full-scale production within a few years.
Robin Deits, a robotics engineer at Boston Dynamics and a PhD from MIT, who has been working there since 2018 focusing on the control of Atlas, gave a detailed presentation on the development journey and lessons learned at the RSS technical sharing event.
In his talk, Deits highlighted the application of MPC (Model Predictive Control) technology in Atlas. MPC is an advanced control strategy capable of handling multi-variable systems and considering constraints, serving as the core technology for Atlas to perform various maneuvers. Deits noted that all Atlas movements since 2019 have been realized through MPC, including parkour, gymnastics, dancing, and backflips.
The MPC version of Atlas shares several common characteristics, including the integration of non-linear dynamics, iterative linearization and QP (Quadratic Programming) solution, not seeking convergence, leveraging problem structure to speed up, and not treating the solver as a black box. Deits and his team have already solved numerous QP problems on the robot and more in simulation environments, ensuring that Atlas can provide robust control solutions in complex environments.
Although Atlas has made significant technical progress, it may still fall over. Engineers learn from these failures and continuously optimize control strategies to enhance the stability and reliability of the robots. As technology continues to advance, the future applications of Atlas and other humanoid robots are promising.
【来源】https://www.jiqizhixin.com/articles/2024-08-18-6
Views: 1