Robust Image-Based Visual Servoing Formation Control for Quadrotors Without Communication via Reinforcement Learning

The image-based visual servoing (IBVS) formation control problem for quadrotors in global navigation satellite system (GNSS)-denied and communication-degraded environments is addressed via reinforcement learning (RL) algorithms. A robust leader-follower formation control method is proposed to achieve formation control, and the follower quadrotors track the leader feature using the virtual camera technique and sphere-based image moments. In order to overcome the absence of intervehicle communication, a leader observer is developed to estimate the state of the leader quadrotor. For the leader-follower IBVS dynamics involving unknown uncertainties and external disturbances, robust visual servoing and attitude control laws are proposed using off-policy RL algorithms without relying on precise dynamical models. The effectiveness of the proposed control scheme is validated through both theoretical analysis and case study results.