Safe and Efficient Control for Dynamic Systems

In this talk I will give an overview of my work to provide scalable safety analysis and control for autonomous systems that must function in the real world among people. I will begin with decomposition techniques for scalable Hamilton-Jacobi reachability analysis, which is a method that provides rigorous safety guarantees but typically suffers from the curse of dimensionality.

I will show how these scalable reachability techniques are applied in the design of FaSTrack: Fast and Safe Tracking, which "robustifies" real-time planning algorithms (e.g. rapidly-exploring random trees, model-predictive control). By precomputing a pursuit-evasion game between a simple model used for online planning and a more realistic model used for tracking, we can determine a maximum tracking error bound between the path generated by the planning algorithm and the actual system. I will show these methods applied to a quadcoptor in a motion capture room planning in real time to navigate around a priori unknown obstacles.

I will then discuss our confidence-aware human prediction framework that similarly allows us to employ simple models of human motion while reasoning about the mismatch between these models and the observed human behavior. This allows us to take advantage of known structure in human motion when it exists, while maintaining safety when this structure is incorrect.

Finally, I will demonstrate these combined methods in both simulation and hardware experiments with multiple humans and multiple quadcoptors navigating through the same room.
Sylvia Herbert is a 5th year PhD student in Electrical Engineering and Computer Science at UC Berkeley. She works with Professor Claire Tomlin in the Hybrid Systems Lab and the Berkeley Artificial Intelligence Research (BAIR) Group. She received her BS and MS in Mechanical Engineering at Drexel University in Philadelphia, PA.