Learning Rollout from Sampling:An R1-Style Tokenized Traffic Simulation Model

1. 1Apply the next-token prediction paradigm to traffic simulation
2. 2Use supervised fine-tuning to iteratively improve the model
3. 3Integrate the proposed R1-style tokenized traffic simulation model to enable active exploration of motion tokens
4. 4Evaluate the model's performance in simulating diverse and high-fidelity traffic scenarios

This research benefits AI engineers and ML researchers working on autonomous driving systems, as it enhances the simulation and evaluation of traffic scenarios

💡 The proposed model improves the active exploration of potentially valuable motion tokens, leading to more realistic traffic simulations

💡 Tokenized traffic simulation model learns rollout from sampling, enhancing autonomous driving evaluation