An Uncertainty-Responsive Safe MPC for Autonomous Driving in Dynamic Environments

This paper presents an uncertainty-responsive model predictive control (MPC) framework designed to ensure the safe operation of autonomous vehicles (AVs) in those environments like, e.g., urban traffic, where the predicted behavior of the surrounding road users (RUs) could be highly uncertain, thus leading to potential collisions between AV and RUs or an unacceptably over-conservative behavior of the AV.In our framework, the collision-avoidance constraints are adjusted on-line to adapt the AV’s behavior to the varying uncertainty of the RUs’ prediction model, such that safety is preserved.Simulations of urban and highway driving scenarios, constructed upon real-world data, show that the proposed approach avoids collisions in presence of unpredicted behaviors of the surrounding RUs.