AgathiKari/act_etiq_pick3_v1

ACT Policy for Tube Pick-and-Place

This is an Action Chunking with Transformers (ACT) policy trained with LeRobot on a custom dataset for picking up a tube and returning to the starting position.

Model Details

• Developed by: Your Name
• Model type: ACT (Action Chunking with Transformers)
• Language(s): English (task instruction)
• License: Apache 2.0
• Finetuned from: ACT (trained from scratch)
• Repository: YOUR_CORRECT_USERNAME/act_etiq_pick3_v1

Uses

Direct Use

This policy can be used with a SO-100 follower robot arm to perform the task: "Grasp the tube, lift it, place it back down, and return to the starting position."

Out-of-Scope Use

Not intended for other tasks or robot morphologies without retraining.

Training Details

Training Data

The model was trained on the YOUR_CORRECT_USERNAME/etiq_pick3 dataset [citation:11], which contains 62 episodes of teleoperated demonstrations. The observation includes a top camera view with bounding-box annotations highlighting the target tube.

Training Procedure

• Framework: LeRobot
• Policy type: ACT
• Chunk size: 20
• Action steps: 10
• Batch size: 8
• Training steps: 100,000 (final checkpoint at step 15K was performing well)
• VAE enabled: Yes

Evaluation

Add your evaluation results here.

How to Use

from lerobot.policies.act import ACTPolicy

policy = ACTPolicy.from_pretrained("YOUR_CORRECT_USERNAME/act_etiq_pick3_v1")