Exploring Heterogeneous Multi-agent Reinforcement Learning with Knowledge Transfer for CACC of ICVs Platoon

Intelligent connected vehicles (ICVs) platoon can be abstracted as a multi-agent system (MAS), and the cooperative adaptive cruise control (CACC) for it has long been a research hot spot. In this paper, a multi-agent reinforcement learning (MARL) with knowledge transfer approach has been developed for solving CACC. Based on the advanced successor feature, the knowledge transfer from source MASs is supposed to enable the control strategy to fully understand the cooperative relationships within various heterogeneous MASs, which is also proved to help the algorithm achieve jump-start improvement. Experimental results demonstrate the effectiveness of the MARL-based CACC solution.