Transferring Human Motion Skills to Humanoid Robots
Transferring human locomotion skills to humanoid robots is a multidisciplinary research field that combines biomechanics, machine learning, and neuroscience. We aim to leverage the neural mechanisms of human locomotion to develop effective locomotion controllers for humanoid robots that can replicate human-like movements and maintain balance [1]. For decades, researchers aim to enable humanoid robots to achieve natural whole-body locomotion patterns and seamless transitions by imitating human motions [2?]. Transferring human motion skills to humanoid robots poses numerous longstanding challenges,
- Describe a tranditional approach? (Use a reference)
Despite significant progress in this area, there are still several challenges to be addressed.
- Summarize the issues (of the traditional approach), e.g.,
(issue 1): Significant morphological differences between human demonstrators and humanoid robots (issue 2):
Additionally, it is not yet clear how to map vision-captured human motion to the robot’s kinematic structure, and how to effectively apply machine learning techniques, such as reinforcement learning, to teach the robot how to walk or perform other locomotion tasks. Further research is needed to overcome these challenges and achieve reliable transfer of human locomotion skills to humanoid robots.
and (issue 3): ? hard to track the human motion with a humanoid? (Link to our other topics?)
- Solutions:
(Regarding issue 1), Our invited speaker, i.e., XXX, recently proposed combining Reinforcement Learning with an adversarial critic component to guide the control policy, ensuring behaviors align with the data distribution of reference motions. Evaluation on a full-sized humanoid JAXON in the simulator showcased a diverse range of locomotion patterns such as standing, push-recovery, squat walking, human-like straight-leg walking, and dynamic running.
Notably, even in the absence of transition motions in the demonstration dataset, the robots exhibit an emerging ability to naturally transition between distinct locomotion patterns as desired speed changes.
(Regarding issue 2), our topics on optimal control, …..
(Regarding issue 3),
Summarize: With deep insights of human motion estimation, and transfer learning, we expect impactful discussions that can advance the state-of-the-art and disseminate knowledge in the international community.
References:
[1] Mombaur, Katja, Anh Truong, and Jean-Paul Laumond. “From human to humanoid locomotion—an inverse optimal control approach.” Autonomous robots 28 (2010): 369-383.