A Model-Free Leader-Follower Approach with Multi-Level Reference Command Generators

This paper introduces an innovative approach to addressing leader-follower control challenges through iterative learning techniques. In this scheme, both the leader and the follower are guided by independent reference generators, simultaneously influencing both entities. The follower is directed by a combination of trajectories, incorporating both the leader's output and its own command generator. The interaction between leader and follower dynamics is captured through a performance index that integrates model-following errors from both entities, thereby shaping the leader's control strategy. Conversely, the follower's performance measure focuses exclusively on its local model-following errors to formulate its control strategy. This method aims to overcome limitations observed in conventional iterative learning control methods, particularly by offering causal strategies based on model-following error dynamics and by explicitly accommodating reference command signals. Notably, this development is achieved within a model-free, data-driven framework, eliminating the need for prior knowledge about the leader-follower system dynamics. Furthermore, the proposed strategies demonstrate flexibility regarding the order of model-following error dynamics. This solution is validated using a heterogeneous system of vehicles characterized by state and control signal delays.