This is a hybrid event with in-person attendance in Wu and Chen and virtual attendance via Zoom. This week’s presenter will be in-person as well. 

ABSTRACT

Deep learning has resulted in remarkable breakthroughs in fields such as speech recognition, computer vision, natural language processing, and protein structure prediction. Robotics has proved to be much more challenging as there are no pre-existing repositories of behavior to draw upon; rather the robot has to learn from its own trial and error in its own specific body, and it has to generalize and adapt. To make this feasible, we have developed “Rapid Motor Adaptation”, a novel technique for adaptive control in the framework of deep reinforcement learning. Using this, we can train robots in simulation and then transfer the skills directly to robots in the real world. I will show multiple examples – quadruped legged locomotion, biped locomotion, in-hand rotation, flying quadcopters – of the success of this approach. I will also show examples of life-long learning in robotics, by continuous adaptation of perception and action in deployed systems.