HChallenge: policy and world action model weights
Code and results: https://github.com/mi-robotics/HChallenge · https://mi-robotics.github.io/HChallenge/
Laid out like the repository's outputs/ folder (scripts/download_weights.sh puts them there):
s1/ablation/full_wrist_state_no_img_pos/ckpt_20000.pt flow-conditioned policy: wrist camera + robot state + action flow
wam/v1/lora_20000.pt world action model v1: LoRA weights for Cosmos-Predict2-2B Video2World
wam/v1/config.json its training config and flow normalisation
wam/prompt_t5.pt T5 embedding of the prompt "pick up the mug and put it on the plate"
The WAM needs the base model nvidia/Cosmos-Predict2-2B-Video2World (gated: accept its licence on Hugging Face).