XuShenLZ/ParkSim

Implements multimodal intent and trajectory prediction using CNN-Transformer architectures for parking lot vehicles, with pre-trained models available for evaluation. The monorepo includes a fleet parking simulator supporting connected autonomous vehicle control, multi-agent coordination, and parking assignment, built on PyTorch Lightning and integrated with the DLP dataset for real-world parking lot trajectories. Provides both prediction pipelines and rule-based simulation environments configurable via YAML for testing vehicle behavior in realistic parking scenarios.