Safe planning using mixed-integer programming for autonomous vehicles coordination

Abstract

Coordination between intelligent vehicles is essential for the advancement of fully autonomous driving. Ensuring safety is the primary focus of this challenge. This paper proposes a safe model predictive planner (MPP) for vehicle coordination, which is robust against uncertainty and noise. The planner provides a feasible route, taking into account the tube of possible trajectories of neighboring vehicles. To achieve this objective, a linear parameter-varying (LPV) prediction model of the vehicle is used. For obstacle avoidance and overtaking maneuvers, mixed-integer linear inequalities as constraints in the MPP formulation are added. Regarding uncertainty and noise, both are assumed to be unknown but bounded and zonotopes are used to enclose and propagate them. Similarly, a zonotopic optimal filter compensates for the measurement noises and estimates the lateral velocity not provided by the vehicle’s instrumentation. The proposed coordination approach is evaluated in a simulation environment, specifically in an aggressive regime with maximum velocity, using $1/10$ scale electric cars.