Events
Seminar @ Cornell Tech: Somil Bansal
Safe and Data-efficient Learning for Robotics
Machine learning has led to tremendous progress in domains such as computer vision, speech recognition, and natural language processing. Fueled by these advances, machine learning approaches are now being explored to develop intelligent physical systems that can operate reliably in unpredictable environments. These include not only robotic systems such as autonomous cars and drones, but also large-scale cyberphysical systems such as transportation and energy systems. However, learning techniques widely used today are extremely data inefficient, making it challenging to apply them to real-world physical systems.
Moreover, they lack the necessary mathematical framework to provide guarantees on correctness, causing safety concerns as data-driven physical systems are integrated in our society. We combine tools from robust optimal control theory with machine learning and computer vision to develop data-efficient and provably safe learning-based control algorithms for physical robotic systems. In particular, we design modular architectures that combine system dynamics models with modern learning-based perception approaches to solve challenging perception and control problems in a priori unknown environments in a data-efficient fashion. Moreover, due to their modularity, these architectures are amenable to simulation-to-real transfer, and can be used for different robotic systems without any retraining. Crucially, we use models not only for faster learning, but also to monitor and recognize the learning system’s failures, and to provide online corrective safe actions when necessary. This allows us to provide safety assurances for learning-enabled systems in unknown and human-centric environments, which has remained a challenge to date.
Speaker Bio
Somil Bansal completed his B.Tech. in Electrical Engineering from Indian Institute of Technology, Kanpur in 2012, and an M.S. in Electrical Engineering and Computer Sciences from UC Berkeley in 2014. Since 2015, he is pursuing a PhD degree in Electrical Engineering and Computer Sciences at UC Berkeley, under the supervision of Prof. Claire Tomlin in the Hybrid Systems Laboratory. His research interests are in exploring how machine learning tools can be combined with the control theoretic frameworks to develop data-efficient and safe learning-based control algorithms for physical robotic systems, especially when the system is operating in an uncertain environment. During his PhD, he has also worked closely with companies like Skydio, Google, Boeing, as well as NASA Ames. Somil has received several awards, most notably the outstanding graduate.