Mechanical and Civil Engineering Seminar

The worlds largest technology companies and science funding agencies are investing heavily in robotics. They anticipate robots that perform work as first responders, efficiently explore the surfaces of planets, and streamline product manufacturing and delivery. However, despite the existence of incredibly capable hardware, the limitations of our best software for controlling and analyzing complex systems prevents us from unleashing these robots into the wild. In this talk, I will describe my research on designing optimization algorithms that improve our ability to control dynamic motions in complex robots. I will present my work developing convex optimization-based controllers for whole-body locomotion and their application to Atlas, a full-scale hydraulic humanoid robot. I will also briefly discuss results from my work developing statistical optimization algorithms for performing risk-sensitive policy search on robots that recover from impacts and manipulate dynamic objects. I will conclude with directions for future research, including adaptive and robust control for low-precision robots and achieving highly dynamic, versatile, and energy-efficient behaviors legged systems.