Reinforcement learning-based funnel control and privacy preservation for multi-agent systems with input dead-zone.

This paper investigates the privacy-preserving protocol and reinforcement learning-based funnel controller design of multi-agent systems subject to input dead-zone constraints. An adaptive funnel controller is formulated to guarantee that the tracking errors keep within prescribed boundaries. The uncharacterized system nonlinearities are approximated by a fuzzy function embedded in an actor-critic reinforcement learning paradigm. To address input constraints and alleviate communication burden, an event-triggered scheme is introduced to update control signals efficiently. Additionally, a secure data-exchange mechanism in light of Paillier cryptographic scheme is designed to safeguard the privacy of state information during transmission. Two comprehensive simulations are performed to validate the feasibility and performance of the developed strategy.