Jaber628/LD4WAM_robotwin

LD4WAM — RoboTwin 2.0 weights

World Dynamics Action Model (WDAM) weights for the RoboTwin 2.0 benchmark from LD4WAM: Learning Latent Dynamics from Human Videos for World Action Models: all 50 tasks (clean_50 + randomized_500 variants), 20-D end-effector actions with proprioception, 3-view observation canvas. The model couples the Wan2.2-TI2V-5B video expert, a latent-dynamics expert supervised by the frozen LDM (Jaber628/LD4WAM_LDM) and an action expert. Code: LD4WAM repository (WDAM/).

Files

File	Description
checkpoint_step_9890.safetensors	full model weights (bf16)
config.yaml	model / dataloader / training config; replace the <...> placeholders with local paths
normalization_stats.npy	min-max action / state normalization statistics (read by deployment)
tokenizer/	UMT5 tokenizer used by the Wan text encoder (kept with the checkpoint so deployment is self-contained)

Usage

Serve the checkpoint as a policy and evaluate on RoboTwin 2.0:

cd LD4WAM/WDAM
bash scripts/deploy.sh /path/to/LD4WAM_robotwin --port 8848 --denoise-steps 10
cd benchmarks/robotwin && bash single_eval.sh adjust_bottle demo_clean wdam 0 8848 127.0.0.1

Or continue training from it:

NPROC_PER_NODE=8 bash scripts/train.sh \
    training.finetune_ckpt_path=/path/to/LD4WAM_robotwin \
    model.video_backbone.model_path=/path/to/Wan2.2-TI2V-5B \
    model.architecture.latent_action_model.ckpt=/path/to/ldm_model_final.pt \
    dataloader.dataset_dir=/path/to/RoboTwin2.0/dataset

See the repository README for environment setup, evaluation and the full option list.

Citation

@misc{shen2026ld4wamlearninglatentdynamics,
      title={LD4WAM: Learning Latent Dynamics from Human Videos for World Action Models},
      author={Zhenhao Shen and Jiaqi Liang and Jasper Lu and Feng Jiang and Yuran Wang and Chuanbo Wei and Jiayi Liu and Jianchun Yang and Qize Yu and Jiadi You and Ce Hao and Guanqi He and Chen Xie and Ruihai Wu},
      year={2026},
      eprint={2608.22403},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2608.22403},
}