PACE-V2X: Planning-aware and communication-efficient semantic interaction for V2X cooperative end-to-end autonomous driving

This PACER-V2X replication package accompanies the PACE-V2X manuscript, providing code, configs, and secondary results. Built on UniV2X, it integrates SACG, TSMF, and PGDP modules. Data: Uses V2X-Seq-SPD (obtain separately from DAIR-V2X-Seq; preprocess per docs).Env: Linux, Python 3.8, CUDA 11.1, PyTorch 1.9.1 (see requirements.txt).Main Exp: Threshold=0.95. Run tools/run_bev_downlink_fused_formal675.sh to evaluate 675 frames for L2 error, collision rate, and communication payload.Training: Run run_bev_downlink_label_and_train.sh for gate training, then calibrate via calibrate_bev_downlink_gates.py.Ablations: Scripts and CSVs included for component ablations, threshold sweeps, and robustness tests.Verification: Validate code against SHA256SUMS.txt. Cite the manuscript and original datasets when using.