train_800_complex__mask__overlay_a75__sim__agentview_camera__live__pi05__seed_0

Model Card for pi05

π₀.₅ (Pi05) is a Vision-Language-Action model from Physical Intelligence designed for open-world generalization: it evolves π₀ to generalize to entirely new environments and situations that were never seen during training. The LeRobot implementation is adapted from their open-source OpenPI repository.

This policy has been trained and pushed to the Hub using LeRobot.

Learn how to train and run it in the LeRobot pi05 guide, or browse the full documentation.

Model Details

• License: apache-2.0
• Fine-tuned from: lerobot/pi05_base
• Robot type: Panda
• Cameras: agentview, robot0_eye_in_hand, robot0_eye_in_hand_2

Inputs & Outputs

The policy consumes these observation features and produces these action features.

Inputs

Outputs

Training Dataset

• Repository: mim-chess-vlas/train_800_complex__mask__overlay_a75__sim__agentview_camera__live
• Episodes: 790
• Frames: 157904
• Frame rate: 20 FPS
• Task(s): "Pick the smooth ball clasped by two flat curved arms that arch over it from one side and almost meet, leaving an open gap between each arm and the ball and place it into the box", "Pick the upright egg flanked by a pair of tall flat ribs on each side, with a chain of small beads curving around its base and place it into the box", "Pick the small even-proportioned shape, with 4 lobes standing out around it, its surface gently textured, cut by deep recesses, widest right at the crown and place it into the box", "Pick the small even-proportioned shape, with 4 lobes standing out around it, its surface smooth and unbroken, more open frame than solid body, pierced by a few openings, widest right at the crown and place it into the box", "Pick the plain smooth cone, circular at the base and tapering to a rounded point, featureless apart from faint rings on the underside and place it into the box", "Pick the ball with one thick arm hooping right over its top and touching down on the far side, with a single stubby foot below and place it into the box", "Pick the smooth four-pointed star, its arms swept down like a starfish, with a small ring-shaped boss at the centre of its top and place it into the box", "Pick the small even-proportioned shape, round in plan, its surface deeply ribbed and knobbly, cut by deep recesses, pierced by a few openings and place it into the box", "Pick the plain smooth pillow-shaped block with one small ball stuck to a face and a shallow dent beside it and place it into the box", "Pick the medium-sized even-proportioned shape, with 4 lobes standing out around it, its surface ribbed, cut by deep recesses and place it into the box", "Pick the small even-proportioned shape, pinched into two lobes, its surface gently textured, more open frame than solid body, widest low down near the base and place it into the box", "Pick the rounded block with one long curved horn arching up and over from its top, and a shallow dimple sunk into one face and place it into the box", "Pick the smooth rounded body with one broad thin blade curving up off it like a sail, and a small dimple on the body below and place it into the box", "Pick the smooth crown of four tall tapering prongs rising from a squat cup, with narrow slots pierced through the cup wall between them and place it into the box", "Pick the smooth rounded cube with a single deep straight slot sawn down into it from the top, splitting one half in two and place it into the box", "Pick the medium-sized squat shape, pinched into two lobes, its surface smooth and unbroken, solid with shallow recesses and place it into the box", "Pick the medium-sized squat shape, with 4 lobes standing out around it, its surface gently textured, more open frame than solid body, pierced by many openings, widest right at the crown and place it into the box", "Pick the small squat shape, round in plan, its surface gently textured, solid with shallow recesses and place it into the box", "Pick the small squat shape, with 4 lobes standing out around it, its surface ribbed, cut by deep recesses, pierced by many openings and place it into the box", "Pick the medium-sized even-proportioned shape, with 4 lobes standing out around it, its surface gently textured, more open frame than solid body, pierced by many openings, widest low down near the base and place it into the box", "Pick the barrel ringed by about eight evenly spaced fat vertical ribs, like a cog, with a clean smooth ring around an open hole at the top and place it into the box", "Pick the clean square box with one round hole in its top face and a small handle bar bridging the hole and place it into the box", "Pick the small even-proportioned shape, with 4 lobes standing out around it, its surface ribbed, cut by deep recesses, pierced by a few openings, widest right at the crown and place it into the box", "Pick the smooth body with four horns curving up and out from its shoulders like a crown, standing on a low ribbed collar and place it into the box", "Pick the small bowl with three thick prongs curling up from its rim around a single bead sitting in the middle, on a ribbed foot and place it into the box", "Pick the plain smooth barrel, egg-like and featureless, with no arms, holes or ridges anywhere and place it into the box", "Pick the faceted pebble, smooth and plain on one face and hollowed by one big open socket on the opposite face, like a scooped-out stone and place it into the box", "Pick the deep round cup with a section of its wall cut clean away, standing on two long arms that reach out sideways and place it into the box", "Pick the thick-walled shallow bowl, plain outside, with one wide round well hollowed into it and a rounded underside and place it into the box", "Pick the open skeletal frame of thin curved struts arching over a shallow saddle, like a rocking cradle with a handle bar above it and place it into the box", "Pick the ball held by two broad straps that rise from the floor on opposite sides and meet in a small loop above it and place it into the box", "Pick the ball wearing a single broad flat flange across its top like a beret, with two short stubby feet under one side and place it into the box", "Pick the smooth broad scoop, like a cupped hand, with three tapering horns rising off its far rim and place it into the box", "Pick the medium-sized even-proportioned shape, with 4 lobes standing out around it, its surface smooth and unbroken, cut by deep recesses, widest low down near the base and place it into the box", "Pick the medium-sized squat shape, with 4 lobes standing out around it, its surface gently textured, more open frame than solid body, pierced by a few openings, widest right at the crown and place it into the box", "Pick the small even-proportioned shape, round in plan, its surface gently textured, solid all through, pierced by a few openings and place it into the box", "Pick the medium-sized even-proportioned shape, round in plan, its surface gently textured, solid with shallow recesses, pierced by a few openings and place it into the box", "Pick the small even-proportioned shape, round in plan, its surface ribbed, more open frame than solid body, pierced by a few openings and place it into the box", "Pick the medium-sized squat shape, with 4 lobes standing out around it, its surface ribbed, cut by deep recesses, pierced by many openings, widest low down near the base and place it into the box", "Pick the medium-sized even-proportioned shape, pinched into two lobes, its surface gently textured, more open frame than solid body, pierced by many openings, widest low down near the base and place it into the box"

Training Configuration

How to Get Started with the Model

New to LeRobot? These guides cover the full workflow:

• Install LeRobot — set up the lerobot package.
• Hardware setup — assemble, wire, and calibrate your robot and cameras.
• Record data & train a policy — the end-to-end imitation-learning walkthrough.
• CLI cheat-sheet — quick reference for the lerobot-* commands.

The short version to run and train this policy:

Run the policy on your robot

lerobot-rollout \
  --strategy.type=base \
  --robot.type=Panda \
  --robot.port=<your_robot_port> \
  --robot.cameras="{ <camera_1>: {type: opencv, index_or_path: <index_or_path>, width: 640, height: 480, fps: 30}, <camera_2>: {type: opencv, index_or_path: <index_or_path>, width: 640, height: 480, fps: 30}}" \
  --policy.path=mim-chess-vlas/train_800_complex__mask__overlay_a75__sim__agentview_camera__live__pi05__seed_0 \
  --task="Pick the smooth ball clasped by two flat curved arms that arch over it from one side and almost meet, leaving an open gap between each arm and the ball and place it into the box" \
  --duration=60

Replace the remaining <...> placeholders with your own values: --robot.port and the camera names/indices are specific to your machine, and the camera names must match the observation keys this policy was trained on.

When --strategy.type=base is used the script doesn't record the episodes. Skipping duration will make the policy run indefinitely. For more information look at rollout documentation.

Train your own policy

This policy type is usually fine-tuned from the pretrained base model lerobot/pi05_base:

lerobot-train \
  --dataset.repo_id=${HF_USER}/<dataset> \
  --policy.path=lerobot/pi05_base \
  --output_dir=outputs/train/<policy_repo_id> \
  --job_name=lerobot_training \
  --policy.device=cuda \
  --policy.repo_id=${HF_USER}/<policy_repo_id> \
  --wandb.enable=true

Writes checkpoints to outputs/train/<policy_repo_id>/checkpoints/.

Evaluation

No evaluation results have been provided for this policy yet.

Citation

If you use this policy, please cite the method linked in the description above, along with LeRobot:

@misc{cadene2024lerobot,
    author = {Cadene, Remi and Alibert, Simon and Soare, Alexander and Gallouedec, Quentin and Zouitine, Adil and Palma, Steven and Kooijmans, Pepijn and Aractingi, Michel and Shukor, Mustafa and Aubakirova, Dana and Russi, Martino and Capuano, Francesco and Pascal, Caroline and Choghari, Jade and Moss, Jess and Wolf, Thomas},
    title = {LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch},
    howpublished = "\url{https://github.com/huggingface/lerobot}",
    year = {2024}
}