Project Gallant

This repository contains the official code for the CVPR 2026 paper:

Voxel Grid-based Humanoid Locomotion and Local-navigation across 3D Constrained Terrains

[ArXiv] · [Project page]

Installation

1. Install Active Adaptation
   Follow the instructions in the active-adaptation repository.
2. Install Gallant (editable mode)
   From the root of this repository:

   pip install -e .

3. Register Gallant projects with Active Adaptation
   Run:

   aa-discover-projects

   and enable gallant and gallant_learning when prompted.

Training

The following command launches distributed PPO training for the Gallant humanoid locomotion task (8 GPUs in this example):

scripts/launch_ddp.sh 0,1,2,3,4,5,6,7 train_ppo.py task=G1LidarNew task.num_envs=1024 algo=ppo_gallant

Adjust the GPU list, task, and task.num_envs according to your hardware and experiment configuration.

Playing / Evaluation

We will release detailed commands and evaluation scripts soon. For now, please refer to the paper and project page for qualitative results.

Citation

If you find Gallant useful in your research, please cite:

@article{ben2025gallant,
  title={Gallant: Voxel grid-based humanoid locomotion and local-navigation across 3d constrained terrains},
  author={Ben, Qingwei and Xu, Botian and Li, Kailin and Jia, Feiyu and Zhang, Wentao and Wang, Jingping and Wang, Jingbo and Lin, Dahua and Pang, Jiangmiao},
  journal={arXiv preprint arXiv:2511.14625},
  year={2025}
}