Optimal Trajectory Planning with Collision Avoidance for Autonomous Vehicle Maneuvering
Authors:
Jason Zalev
Abstract:
To perform autonomous driving maneuvers, such as parallel or perpendicular parking, a vehicle requires continual speed and steering adjustments to follow a generated path. In consequence, the path's quality is a limiting factor of the vehicle maneuver's performance. While most path planning approaches include finding a collision-free route, optimal trajectory planning involves solving the best tra…
▽ More
To perform autonomous driving maneuvers, such as parallel or perpendicular parking, a vehicle requires continual speed and steering adjustments to follow a generated path. In consequence, the path's quality is a limiting factor of the vehicle maneuver's performance. While most path planning approaches include finding a collision-free route, optimal trajectory planning involves solving the best transition from initial to final states, minimizing the action over all paths permitted by a kinematic model. Here we propose a novel method based on sequential convex optimization, which permits flexible and efficient optimal trajectory generation. The objective is to achieve the fastest time, shortest distance, and fewest number of path segments to satisfy motion requirements, while avoiding sensor blind-spots. In our approach, vehicle kinematics are represented by a discretized Dubins model. To avoid collisions, each waypoint is constrained by linear inequalities representing closest distance of obstacles to a polygon specifying the vehicle's extent. To promote smooth and valid trajectories, the solved kinematic state and control variables are constrained and regularized by penalty terms in the model's cost function, which enforces physical restrictions including limits for steering angle, acceleration and speed. In this paper, we analyze trajectories obtained for several parking scenarios. Results demonstrate efficient and collision-free motion generated by the proposed technique.
△ Less
Submitted 13 May, 2025;
originally announced May 2025.
