Expressive Bipedal Robotic Character with Reinforcement Learning

Here is an expressive bipedal robotic character with a reinforcement learning-based control architecture that can execute artistic motions. It is remote controlled with a puppeteering interface which allows for expressive performances to be created on the fly. As the researchers explain:

command signals are generated by an animation engine which composes and blends between multiple animation sources.

This robot uses off-the-shelf hardware modules. It has 5 actuators in each leg and 4 in its neck. It has speakers, illuminated eyes and a headlamp. The above video from Disney Research Hub shows it in action.

[HT]