Co-Active Learning to Adapt Humanoid Movement for Manipulation.

In this paper, we address the problem of interactive robot movement adaptation under various environmental constraints. A common approach is to adopt motion primitives to generate target motions from demonstrations. However, their generalization capability is weak for novel environments.
Additionally, traditional motion generation methods do not consider versatile constraints from different users, tasks, and environments. In this work, we propose a co-active learning framework for learning to adapt the movement of robot end effectors for manipulation tasks. It is designed to adapt the original imitation trajectories, which are learned from demonstrations to novel situations with different constraints. The framework also considers user feedback towards the adapted trajectories, and it learns to adapt movement through human in-the-loop interactions. Experiments on a humanoid platform validate the e effectiveness of our approach.