EmLoco — CVPR 2025 release assets

Checkpoints and preprocessed data for Physical Plausibility-aware Trajectory Prediction via Locomotion Embodiment (CVPR 2025).

📄 paper · 🐙 code

Contents

• checkpoints/{jta_ours, jrdb_ours}/ — Our models trained on JTA/JRDB dataset
• checkpoints/valuenets/ — LocoVal value-network checkpoints used by evaluate_*.py / train_*.py --valueloss_w
• preprocess_smpl_cvpr/{jta, jrdb}_all_visual_cues/{train, val, test}/ — preprocessed data
• action_dict.json — JRDB-Act action labels for per-action ADE/FDE breakdown (optional)

Usage

See the GitHub README for download + setup instructions.

License

CC BY-NC 4.0 (research, non-commercial). The underlying SMPL / JTA / JRDB / PACER restrict commercial use.

Citation

@InProceedings{EmLoco_CVPR25,
  author    = {Taketsugu, Hiromu and Oba, Takeru and Maeda, Takahiro and Nobuhara, Shohei and Ukita, Norimichi},
  title     = {Physical Plausibility-aware Trajectory Prediction via Locomotion Embodiment},
  booktitle = {IEEE / CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2025}
}