Distributed V-formation control via fat-leader MPC, feedforward consensus and soft-feasible ECBF-QP

V-shaped formation control of autonomous ground vehicles (AGVs) is attractive for convoy operation, but obstacle-cluttered environments make formation tracking and safety filtering difficult to satisfy simultaneously. This paper develops a three-layer distributed control framework for a virtual leader and three car-like follower vehicles. The planning layer uses a fat-leader model predictive controller (MPC), in which the leader is assigned a 6.5 m circular planning proxy to bias the reference trajectory away from obstacle corners. The tracking layer uses a body-frame decoupled controller with a rotational feedforward gain of 0.25 to compensate part of the velocity demand created by V-formation turning. The safety layer applies a second-order exponential control barrier function quadratic program (ECBF-QP) with a shared high-penalty slack variable. The softened QP remains feasible, while forward invariance requires zero slack over the closed-loop trajectory and admissible initial higher-order barrier states. In a 100 m by 100 m S-curve simulation with four circular obstacles, the revised implementation maintains the 2.5 m obstacle and inter-vehicle safety thresholds, with minimum obstacle clearance of 3.98 m and minimum inter-vehicle distance of 3.58 m. Component ablations show that removing the ECBF-QP causes an obstacle-safety violation (1.19 m clearance), while removing rotational feedforward increases mean tracking error from 7.50 to 7.98 m. The results support footprint-aware planning and safety filtering in the tested scenario, while exposing the associated tracking-error cost and the conditional nature of the theoretical safety statement.