Rapid motion planning and autonomous obstacle avoidance for unmanned vehicles

摘要

This work introduces the use of optimal control methods for path planning and control of autonomous vehicles in an obstacle-rich environment. Traditional techniques harbor non-optimal, closed architectures primarily derived at a time when computational complexity could significantly hinder overall system performance. Advancements in computing power, miniaturization, and numerical methods permit the utilization of online, optimal path planning and control, thereby improving system flexibility and autonomy. The backbone of this concept is state-of-the-art optimal control techniques involving pseudospectral methods and sequential quadratic programming. Although this research focuses on a robotic car or Unmanned Ground Vehicle (UGV), several systems, including an Unmanned Aerial Vehicle (UAV) and a pendulum on a rotational base, are detailed to illustrating the technique's modularity. With respect to the UGV, optimal control methods permit the optimization of maneuver parameters while accounting for complex vehicle kinematics and workspace obstacles, represented as dynamic and path constraints respectively. The path constraints are modeled such that an obstacle of any shape or size can be included. Maneuvering trajectories are first generated in an open-loop architecture, followed by an application of these same techniques in feedback form. Lastly, model fidelity is increased to improve control over vehicle behavior and closed-loop performance and a local knowledge scenario is evaluated.