pi0

pi0.5

“How can we enable better generalization?” - By training on many surrogate tasks

Training in Lerobot

  python src/lerobot/scripts/lerobot_train.py \
      --dataset.repo_id=Gongsta/koch-tshirt-folding-v3 \
      --dataset.video_backend=pyav \
      --policy.type=pi05 \
      --output_dir=./outputs/pi05_training \
      --job_name=pi05_training \
      --policy.repo_id="Gongsta/pi05-koch-tshirt" \
      --policy.pretrained_path=lerobot/pi05_base \
      --policy.compile_model=true \
      --policy.gradient_checkpointing=true \
      --wandb.enable=true \
      --policy.dtype=bfloat16 \
      --policy.freeze_vision_encoder=false \
      --policy.train_expert_only=false \
      --steps=6000 \
      --save_freq=2000 \
      --policy.device=cuda \
      --batch_size=32 \
      --num_workers=4

python src/lerobot/scripts/lerobot_train.py\
    --dataset.repo_id=Gongsta/koch-tshirt-folding-v3 \
    --policy.type=pi05 \
    --output_dir=./outputs/pi05_training \
    --job_name=pi05_training \
  --policy.repo_id="Gongsta/pi05-koch-tshirt" \
    --policy.pretrained_path=lerobot/pi05_base \
    --policy.compile_model=true \
    --policy.gradient_checkpointing=true \
    --wandb.enable=true \
    --policy.dtype=bfloat16 \
    --policy.freeze_vision_encoder=false \
    --policy.train_expert_only=false \
    --steps=6000 \
    --save_freq=2000 \
    --policy.device=cuda \
    --batch_size=32

Citation

(Intelligence et al., 2025)

Intelligence, P., Black, K., Brown, N., Darpinian, J., Dhabalia, K., Driess, D., Esmail, A., Equi, M., Finn, C., Fusai, N., Galliker, M. Y., Ghosh, D., Groom, L., Hausman, K., Ichter, B., Jakubczak, S., Jones, T., Ke, L., LeBlanc, D., … Zhilinsky, U. (2025). π_0.5: a Vision-Language-Action Model with Open-World Generalization. https://arxiv.org/pdf/2504.16054