juyil/spatial-b8-16act-2token-60ksteps

This repository contains the checkpoint trained on Spatial LIBERO task suite, chunk size 16, Each GPU batch size 8, as described in VOTE: Vision-Language-Action Optimization with Trajectory Ensemble Voting (arXiv).

Official code repository: LukeLIN-web/VOTE

Eval Parameter

num_blocks 2, num_Actions_chunk 16 , --num_actions_per_token", "8" ,   , mode mul,  hidden dim 4096 .

Citation

If you use our code in your work, please cite our paper:

@misc{lin2025votevisionlanguageactionoptimizationtrajectory,
      title={VOTE: Vision-Language-Action Optimization with Trajectory Ensemble Voting}, 
      author={Juyi Lin and Amir Taherin and Arash Akbari and Arman Akbari and Lei Lu and Guangyu Chen and Taskin Padir and Xiaomeng Yang and Weiwei Chen and Yiqian Li and Xue Lin and David Kaeli and Pu Zhao and Yanzhi Wang},
      year={2025},
      eprint={2507.05116},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2507.05116}, 
}