CoIRL-AD: Collaborative-Competitive Imitation-Reinforcement Learning in Latent World Models for Autonomous Driving
Researchers have introduced CoIRL-AD, a novel framework for training autonomous driving models that combines imitation learning (IL) and reinforcement learning (RL) in an offline setting. This approach aims to improve generalization, particularly in rare scenarios, by decoupling IL and RL objectives and using imagined rollouts for reward estimation. Experiments on the nuScenes benchmark demonstrated that CoIRL-AD enhances robustness and cross-city generalization compared to existing IL-based methods. AI