Lin Shao NUS
时间： 2022-12-01 14:00-2022-12-01 15:00
地点：Tencent Meeting ID: 772-848-7233 (https://meeting.tencent.com/p/7728487233)
Model-based reinforcement learning (MBRL) is widely recognized with the potential to be significantly more sample efficient than model-free reinforcement learning. How an accurate model can be developed automatically and efficiently from raw sensory inputs (such as images), especially for complex environments and tasks, is a challenging problem that hinders the broad application of MBRL in the real world. Recent developments in differentiable physics-based simulation and rendering provide a potential direction. In this talk, I will introduce a learning framework called SAGCI that leverages differentiable physics simulation to model the environment. It takes raw point clouds as inputs and produces an initial model represented as a Unified Robot Description Format (URDF) file, which is loaded into the simulation. The robot then utilizes interactive perception to online verify and modify the model. We propose a model-based learning algorithm combining object-centric and robot-centric stages to produce policies to accomplish manipulation tasks. Next, I will present a sensing-aware model-based reinforcement learning system called SAM-RL, combining differentiable physics simulation and rendering. SAM-RL automatically updates the model by comparing the rendered images with real raw images and produces the policy efficiently. With the sensing-aware learning pipeline, SAM-RL allows a robot to select an informative viewpoint to monitor the task process. We apply our framework to real-world experiments for accomplishing three manipulation tasks: robotic assembly, tool manipulation, and deformable object manipulation. I will close this talk by discussing the lessons learned and interesting open questions that remain.
Lin Shao is an Assistant Professor in the Department of Computer Science at the National University of Singapore (NUS), School of Computing. His research interests lie at the intersection of Robotics and Artificial Intelligence. His long-term goal is to build general-purpose robotic systems that intelligently perform a diverse range of tasks in a large variety of environments in the physical world. Specifically, his group is interested in developing algorithms and systems to provide robots with the abilities of perception and manipulation. He is a co-chair of the Technical Committee on Robot Learning in the IEEE Robotics and Automation Society. Previously, he received his Ph.D. at Stanford University and his B.S. at Nanjing University.