Distributed nonlinear MPC formation control with limited bandwidth

We address leader-follower formation control of autonomous vehicles in a non-ideal communication environment, e.g. underwater channel, where the bandwidth is limited and there are communication and computational delays. Moreover, the agents have both input and state constraints on their dynamics. A novel formulation of nonlinear model predictive control (NMPC) is presented, in which agents do not need to estimate neighbors' dynamics and collision avoidance is guaranteed. Packet size is reduced considerably by data compression with neural networks. Moreover, this method allows the agents to be sampled at different rates, have different dynamics, constraints and prediction horizons, and be robust to propagation delays. Collision avoidance is achieved by means of a spatial filter based potential field. The sound analytical results are verified by simulations.