GoGE Workshop on Safe Autonomous Systems
Nov. 12, 10-12:15pm KST (Nov. 11, 5-7:15pm PST) via Zoom
Zoom link: https://snu-ac-kr.zoom.us/j/85306283487
Talk 1. 10-10:45am KST (5-5:45pm PST)
Sylvia Herbert, UC San Diego
Title: Constructing Safety Analyses Efficiently for Hamilton-Jacobi Reachability and Control Barrier Functions
Abstract: Safety analysis tools for nonlinear systems (Hamilton-Jacobi Reachability, Control Barrier Functions) suffer from scalability and generalizability. This is crucial for real-world scenarios, where safe control must be computed for high-dimensional systems operating in uncertain environments. Moreover, these safety analyses must be adaptable to changing knowledge and information about the system and the environments. In this talk I will review some recent work for (a) scaling HJ reachability analysis and (b) syncing theory from HJ reachability analysis and Control Barrier Functions in order to merge benefits of both methods.
Biography: Sylvia Herbert is an Assistant Professor in Mechanical and Aerospace Engineering at UC San Diego. Her research interests are in developing theoretically sound techniques for efficiently guaranteeing safe control based on available models of systems and given information about environments. These techniques should be able to quickly adapt to unexpected changes and new information in the autonomous system or the environment. More information on her research interests can be seen at sylviaherbert.com.
Talk 2. 10:45-11:30am KST (5:45-6:30pm PST)
Somil Bansal, Univ. of Southern California
Title: Scaling Reachability Analysis for Robotics: High-Dimensional Systems to Real-Time Computation
Abstract: Hamilton-Jacobi Reachability is a powerful verification tool for robotic systems. However, it is challenging to scale the reachability analysis to the scenarios where the underlying system is high-dimensional, multiple agents are operating in the same environment, or when a real-time update of the safe set is required, all of which are commonly occurring scenarios in robotics. We present new methods for computing the reachable set, based on a functional approximation that has the potential to broadly alleviate its computational complexity. In the second part of the talk, we will present a toolbox of methods that can leverage previously computed solutions to update the safety guarantees online within a fraction of milliseconds, as new environment information is obtained. We will illustrate these methods on various robotic platforms, including demonstrations of motion planning around people, and navigating in a priori unknown environments.
Biography: Somil Bansal is an Assistant Professor at the Department of Electrical Engineering of the University of Southern California, Los Angeles. He received a Ph.D. in Electrical Engineering and Computer Sciences (EECS) from the University of California at Berkeley in 2020. Before that, he obtained a B.Tech. in Electrical Engineering from the Indian Institute of Technology, Kanpur, and an M.S. in Electrical Engineering and Computer Sciences from UC Berkeley in 2012 and 2014, respectively. Between August 2020 and August 2021, he spent a year as a Research Scientist at Waymo (formerly known as the Google Self-Driving Car project). He has also collaborated closely with companies like Skydio, Google, Waymo, Boeing, as well as NASA Ames. Somil is broadly interested in developing mathematical tools and algorithms for the control and analysis of autonomous systems, with a focus on bridging learning and control-theoretic approaches for safety-critical autonomous systems. Somil has received several awards, most notably the Eli Jury Award at UC Berkeley for his doctoral research, the outstanding graduate student instructor award at UC Berkeley, and the academic excellence award at IIT Kanpur.
Talk 3. 11:30-12:15pm KST (6:30-7:15pm PST)
Astghik Hakobyan, Seoul National University
Title: Risk-Aware Distributionally Robust Optimization for Learning-based Autonomous Systems
Abstract: Motion planning and control are considered to be fundamental problems in today’s robotics as robots are making their way into the cluttered urban environment where humans inhabit. Therefore, a crucial task is constructing a path for the robot and controlling it while considering the dynamic and possibly uncertain behavior of the environment. State-of-the-art learning techniques enable predicting the uncertainties in the perceived information about surroundings, making the operation of the robots more intelligent. The difficulty arises when the environment changes rapidly and predictions about its future realizations are inaccurate. We present our recent works on risk-aware motion planning and control techniques for learning-based autonomous systems, which are robust against distributional uncertainties of the environment.
Biography: Astghik Hakobyan is a Ph.D. student at the Seoul National University (SNU), Seoul, South Korea. She received B.S. degree in Automation and Control from the National Polytechnic University of Armenia (NPUA), Yerevan, Armenia in 2018; and an M.S. degree in Electrical and Computer Engineering (ECE) department from SNU in 2020. Her research interests include control and optimization, motion planning and safe autonomy of robots.
Poster session (Youtube link: https://www.youtube.com/watch?v=QzD8jF6w0Lo)
Yeoneung Kim, Seoul National University
Title: Training Wasserstein GANs without gradient penalties
Acknowledgment: Supported by BK21 Four Education and Research Program for Future ICT Pioneers and the Information and Communications Technology Planning and Evaluation (IITP) grant funded by MSIT(2020-0-00857).