Collision Avoidance for Multiple Static Obstacles using Path-Velocity Decomposition

This paper proposes an algorithm to plan collision-free trajectories for emergency scenarios using path-velocity decomposition (PVD). PVD decouples longitudinal and lateral planning, enabling the use of a linear model predictive control based path planner, which can operate at the bounds of admissible controls, and has low computational cost. However, infeasibility may occur in this method due to insufficient lateral acceleration capability during evasive manoeuvres. The problem is further complicated by the presence of multiple obstacles, which may require multiple evasive manoeuvres. Our approach overcomes this issue by connecting individually feasible evasive manoeuvres using heuristics in order to plan a velocity profile that guarantees collision avoidance throughout the prediction horizon. For this purpose, a simple method to approximate the acceleration profile to maximize deviation for a given obstacle distance is used, which considers actuator as well as friction constraints. These profiles are then used in motion primitives to perform collision checks in the velocity planning stage, thus ensuring feasibility in the path planning stage. Simulation results for critical last-second emergency scenarios consisting of multiple static obstacles show that a collision-free trajectory on friction limits can be planned using our approach. (C) 2019, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.