README.md

metadata

license: cc-by-nc-sa-4.0
language:
  - en
tags:
  - human-trajectory-prediction
  - jrdb
  - jta
  - physics-simulation
  - paa2026
pipeline_tag: other
library_name: pytorch

Hexa-P: PAA 2026 release assets

Checkpoints and preprocessed data for Hexa-P: Physical Plausibility-aware Pre-training for Pedestrian Path Prediction (Pattern Analysis and Applications, 2026).

🐙 code · 🌐 project page

Contents

• weights/stage1/hexap_pretrain.ckpt: Hexa-P pre-trained in the physics simulation (5 hypotheses), the starting point of fine-tuning
• weights/stage1/hexap_pretrain_deterministic.ckpt: the same with one hypothesis
• weights/stage1/controller.ckpt: humanoid controller that walks the pre-training paths
• weights/stage2/{jrdb, jta}/hexap.pt: Hexa-P fine-tuned on JRDB and JTA
• data/{jta, jrdb, vru}/{train, val, test}/: preprocessed data (vru is the Urban dataset)
• data/jta_replay/test/: JTA test scenes for the robot navigation experiment

Usage

See the GitHub README for download + setup instructions.

License

Research, non-commercial use only. data/<dataset>/LICENSE gives the licence of each folder: JTA CC BY-NC 4.0, JRDB CC BY-NC-SA 4.0, Urban CC BY 4.0. The weights were trained with SMPL and AMASS, which restrict commercial use. The JTA authors require the statement: "I declare that I will use the JTA Dataset for research and educational purposes only, since I am aware that commercial use is prohibited. I also undertake to purchase a copy of Grand Theft Auto V."

Citation

@article{taketsugu2026hexap,
  title   = {Hexa-P: Physical Plausibility-aware Pre-training for Pedestrian Path Prediction},
  author  = {Taketsugu, Hiromu and Nobuhara, Shohei and Ukita, Norimichi},
  journal = {Pattern Analysis and Applications},
  year    = {2026}
}