Meta-Multi-Agent Reinforcement Learning for Fast Adaptation of Interactive Policies with Applications to Autonomous Driving
This work introduces Meta-MARL to rapidly adapt policies for multi-agent systems by modeling problems as Markov games and defining a meta-Nash equilibrium (meta-NE). It uses a MAML-style approach based on Markov Potential Games (MPGs) to allow agents to quickly adjust their behavior when faced with different driving styles, showing faster adaptation than standard methods in autonomous driving.