Cross-morphology Imitation (CMI).

In short, this is about learning control policies for non-human robots (for example, legged manipulators such as Spot) from human motion data.

One of the easiest ways to understand this direction is to look at the work of Tianyu Li in our group (https://easypapersniper.github.io/) , which provides concrete examples of what this line of research looks like.

• Your previous research,
• What kind of the immediate next research project you would like to pursue under the theme of CMI (the more concrete, the better), and
• Your longer-term research vision if you were to work on CMI-related topics over the next five years.

Cross Morphology Imitation

1. Imitation

ACE: Adversarial Correspondence Embedding for Cross Morphology Motion Retargeting from Human to Nonhuman Characters

CROSSLOCO: HUMAN MOTION DRIVEN CONTROL OFLEGGED ROBOTS VIA GUIDED UNSUPERVISED REIN-FORCEMENT LEARNING

MOREFLOW: MOTION RETARGETING LEARNING THROUGH UNSUPERVISED FLOW MATCHING

2. Interactions

2-1. Simulation and Retargeting of Complex Multi-Character Interactions