CtRL-Sim leverages return-conditioned offline reinforcement learning to enable the generation of reactive, closed-loop, and controllable driving agent behaviors within a physics-enhanced simulation environment.
