Safe Reinforcement Learning-Based Visual Servoing Control for Quadrotors Tracking Unknown Ground Vehicles

Abstract

The visual servoing control problem with multiple constraints for the quadrotor to track an unknown ground vehicle is addressed via safe reinforcement learning. The tracking control problem for the unknown vehicle in the absence of the global navigation satellite system is transformed into solving a visual servoing control problem for the time-varying system. A Velocity observer is developed to estimate the unknown motion of the ground vehicle, and a visual servoing control law is proposed by a reinforcement learning-based optimal control with an online actor-critic structure and a backstepping-based control. Barrier Lyapunov functions and nonquadratic utility functions are introduced to keep the multiple constrained visual servoing system in the safe sets. The stability of the proposed visual servoing control laws is proven, and simulation results of the quadrotor tracking an unknown ground vehicle are provided to demonstrate the effectiveness of the control laws.