act_Final_Sim
robotics policy
1 min readupdated
A trained robotics control policy built with the LeRobot library. It uses ACT (Action Chunking with Transformers, an imitation learning method that predicts short action chunks instead of individual steps). Trained on teleoperation demonstration data, the model processes visual camera observations alongside the robot's state. It is well-suited for robotics researchers and developers testing autonomous control.
Key features
- Trained with the LeRobot library
- Processes visual observations from an overhead camera
- Designed for so_follower type robots
- Trained on a 15-episode dataset at 30 frames per second
What you can do
- Run the policy on a real robot using LeRobot utilities
- Train your own policy on similar datasets
- Simulate and test autonomous robot actions
Example
Run the robot control policy in your terminal:
lerobot-rollout \
--strategy.type=base \
--robot.type=so_follower \
--robot.port= \
--robot.cameras="{ : {type: opencv, index_or_path: , width: 640, height: 480, fps: 30}}" \
--policy.path=Screener2/act_Final_Sim \
--task="0" \
--duration=60Read more
DetailsLicense: apache-2.0
Training steps: 100,000
Batch size: 8
Optimizer: adamw
Learning rate: 1e-05










Comments