[PAST] Structured Prediction for Sensorimotor Control
[PAST] Structured Prediction for Sensorimotor Control [1,580]
2019 / 01 / 11 AM 11:00
Location: 301 - 1420
Speaker: Patrick Wensing

Patrick Wensing is an Assistant Professor in the Department of Aerospace and Mechanical Engineering at the University of Notre Dame. He received his Ph.D. in Electrical and Computer Engineering from The Ohio State University in 2014, and completed Postdoctoral training at MIT in 2017. He was awarded an NSF Graduate Research Fellowship for his dissertation research on balance control strategies for humanoid robots. At MIT, the results of his postdoctoral work on the MIT Cheetah robot have received considerable publicity worldwide, with features from TIME, WIRED, and the Wall Street Journal.


Recent technological advances have given way to a new generation of versatile legged robots.  These machines are envisioned to replace first responders in disaster scenarios and enable unmanned exploration of distant planets.  As our robots are sent into these environments, however, they must include new capacities for reasoning through the complex sequences of actions necessary to negotiate challenging terrains.  

Toward this aim, this talk details the development of control systems for legged robots to achieve unprecedented levels of dynamic mobility through structured prediction.  Drawing on key insights from biomechanics, the talk will open with a description of optimization-based balance control algorithms for high-speed locomotion in humanoid robots.  It will then present control algorithms for the MIT Cheetah 2 quadruped robot that enable dynamic locomotion in experimental hardware.  A model predictive control framework for this robot will be described, which enables the Cheetah to autonomously jump over obstacles with a maximum height of 40 cm (80% of leg length) while running at 2.5 m/s.  The talk will include discussion of extensions to this framework for 3D balance control with a new experimental quadruped, the MIT Cheetah 3.  Throughout, principled reductions will be shown to admit structure for the problem of predictive control, critically enabling real-time application. A few speculative remarks will be provided regarding the use of similar abstractions for the dual problem of estimation within the context of human-robot interaction.