September 27, 2024, Webb Hall 1100

Learning-based planning and control for robotics promises to handle complex dynamics, sensors, and tasks with less engineering effort than previous methods. However, many impressive empirical results are brittle: both the learned policy and the learning algorithm itself may fail under seemingly minor changes in the setting. To realize the full potential of robot learning, we must make these algorithms reliable.In this talk, I will share work towards this goal in both theory and practice. First, I will present GAPS, a principled and general method for optimizing a control policy when both dynamics and costs are time-varying and revealed online. GAPS enjoys minimax regret optimality in simple settings and a novel "local regret" guarantee in nonconvex settings, and also demonstrates strong practical performance in our hardware experiments with quadrotors and cars. Then, I will demonstrate the benefits of combining the expressive power of deep dynamics models with the interpretability of nonlinear control for 1) complex partially-observable tasks such as deformable object manipulation, and 2) adaptive ground vehicle navigation on challenging terrains via visual foundation models.Finally, I will outline research plans towards a unified framework for reliable robot learning. These plans are centered around investigating the structural properties that make learning-to-control tractable in physical systems, while keeping a tight connection between theory and practice.

James A. Preiss is an Assistant Professor in the Computer Science Department at the University of California, Santa Barbara. Previously, he obtained a Ph.D. in Computer Science from the University of Southern California and was a postdoctoral scholar at Caltech. He has made contributions to a broad spectrum of topics in robotics, including the Crazyswarm software platform for multi-quadrotor research. His current interests focus on building rigorous and practical methods for learning-based planning and control, with diverse real-hardware deployments