A Non-Optimization-Based Dynamic Path Planning for Autonomous Obstacle Avoidance

This paper presents a non-optimization-based frame- work for path planning and tracking for evasive maneuvers in autonomous cars. The framework exploits a two-layer approach where a path planner generates a reference trajectory that is then tracked by a path tracking controller. A nested curvature preview controller implements path tracking. In the paper, we show how to describe the closed-loop performance of the controller. The quantification of the closed-loop performance in the frequency domain guides the generation of the evasive path. In this way, the algorithm generates a path that avoids the obstacle (if possible) accounting for both static and dynamic constraints. The proposed framework thus provides a non-optimization-based way to integrate the characteristics of the path tracker in the path planner algorithm, thus avoiding the need to define cost functions and use third party optimizers. The paper validates the proposed evasive maneuver strategy in simulation and on an instrumented vehicle. First, we test the trajectory tracker, showing that it tracks aggressive trajectories (with lateral acceleration close to 1 g) with an error smaller than 30 cm. Subsequently, we integrate the curvature preview with the path generator and show the joint generation-tracking performance in two different scenarios.