Two of our papers have been accepted at CoRL2021, for poster presentations November 8-11. Learn more by checking out the pre-prints below, or contacting the authors (Yue and Charles are happy to talk through any questions you may have).
"Safe Nonlinear Control Using Robust Neural Lyapunov-Barrier Functions" by Charles Dawson, Zengyi Qin, Sicun Gao, and Chuchu Fan (link)
This paper addresses questions about how to build reliable, trustworthy control systems out of neural networks, while providing guarantees on how well those controllers will generalize to systems with uncertain model parameters (like changed mass or damping, or changed aerodynamic properties).
"Learning Density Distribution of Reachable States for Autonomous Systems" by Yue Meng, Dawei Sun, Zeng Qiu, Md Tawhid Bin Waez, Chuchu Fan (link)
This paper provides an efficient method for calculating the likelihood that an autonomous system will reach some region, using neural networks to learn the future density distribution of the system. This could allow system designers to easily calculate, for example, the probability of colliding with an obstacle or reaching the goal, as well as provide online monitoring for system safety.