IPAB Workshop-02/07/2020
Title: Learning natural locomotion behaviours for humanoid robots using human bias Abstract: In this talk, I will present a learning framework that bridges knowledge from imitation learning, deep reinforcement learning, and control theories to achieve human-style locomotion of humanoid robots that is natural, dynamic, and robust. I proposed novel approaches to introduce human bias, ie motion capture data and a special Multi-Expert network structure. I used the Multi-Expert network structure to smoothly blend behavioral features, and used the augmented reward design for the task and imitation rewards. The reward design is more composable, tunable, and explainable by using fundamental concepts from conventional humanoid control. I rigorously tested and benchmarked the learning framework on different disturbance scenarios and consistently produced robust locomotion behaviors. Further, I demonstrated the capability of learning robust and versatile policies in presence of disturbances, such as terrain
Title: Synthesis of Detailed Hand Manipulations
Abstract: We have captured a wide range of hand-object manipulation motions. Using this data, we are developing a deep-learning framework to synthesize the finger motion for manipulating various objects. We propose to utilize different distance information sensed between the hand and the object as the geometry representation. We show such representation could help the model have a better generalization even training with few shapes.
IPAB Workshop-02/07/2020
Blackboard Collaborate https://eu.bbcollab.com/guest/ed8634d969604610922e7a0200b1a3b9