Dictionary entry Trending
cross-embodiment learning
also cross-embodiment or cross-embodiment transfer or multi-embodiment learning
Definition of cross-embodiment learning
-
: robot learning that transfers data, representations, skills, or policies between bodies with different sensors, shapes, joints, or action spaces, such as from a person to a robot or from one robot type to another
- Cross-embodiment learning allowed the policy to benefit from demonstrations collected with several different robot arms.
- The team used human videos for cross-embodiment transfer instead of recording every task again on the target humanoid.
Origin & history
Cross-embodiment appeared as a robotics research label by 2021 in work on transferring skills between human and robot bodies. The field gained wider visibility in 2023 through Google DeepMind and the Open X-Embodiment collaboration, which pooled demonstrations from many robot types to train policies that could benefit from experience gathered on other embodiments.
Editorial references
Sources & further reading
- XIRL: Cross-embodiment Inverse Reinforcement Learning↗ Research Paper · arXiv · 2021-06-07
- Scaling up learning across many different robot types↗ Research Announcement · Google DeepMind · 2023-10-03
- Intelligence Across Embodiments↗ Research Perspective · arXiv · 2026-09-22
Test yourself
Which of these is the meaning of cross-embodiment learning?
Cite this entry
"Cross-embodiment learning." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/cross-embodiment-learning/. Accessed 16 Oct. 2026.