shraavb/act_pick_place_cube

ACT Policy for Pick and Place Cube Task

This model was trained using LeRobot to perform a pick-and-place task with an SO-101 robot arm.

Task Description

Task: Pick up a cube and place it in a container

The robot learns to:

1. Locate the cube visually
2. Move the gripper to the cube
3. Grasp the cube
4. Transport it to the target container
5. Release the cube into the container

Model Details

• Policy Type: ACT (Action Chunking with Transformers)
• Robot: SO-101 Follower Arm (6-DOF with gripper)
• Camera: Single front-facing RGB camera (640x480 @ 30fps)
• Training Steps: 100,000
• Training Episodes: 51 demonstrations
• Framework: LeRobot

Training Configuration

policy:
  type: act
  chunk_size: 100
  n_action_steps: 100
  dim_model: 512
  n_heads: 8
  n_encoder_layers: 4
  n_decoder_layers: 1
  vision_backbone: resnet18
  use_vae: true

Usage

Evaluation on Robot

lerobot-record \
  --robot.type=so101_follower \
  --robot.port=/dev/ttyACM0 \
  --robot.cameras='{ front: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30} }' \
  --dataset.repo_id=YOUR_USERNAME/eval_pick_place_cube \
  --dataset.single_task="Pick up the cube and place it in the container" \
  --policy.path=shraavb/act_pick_place_cube

Training Dataset

• Dataset: shraavb/pick_place_cube
• Episodes: 51
• Frames: ~45,000

Hardware Requirements

• SO-101 robot arm with Feetech STS3215 servos
• USB camera (webcam or similar)
• Computer with USB ports for robot and camera

Acknowledgments

Trained following the Seeed Studio LeRobot tutorial.