lewam_jepa_pusht
robotics policy
1 min readupdated
A new robotics policy trained using the LeRobot library for block-pushing tasks. The model takes camera images and robot state as input and outputs control actions. It is trained on the pusht demonstration dataset and can run inference on real hardware. This solution is useful for researchers and developers experimenting with imitation learning algorithms in robotics.
Key features
- Trained using the LeRobot framework
- Specialized in pushing a T-shaped block
- Based on a dataset with a 10 FPS frequency
- Uses the adamw optimizer
What you can do
- Run the policy on your own robot using LeRobot CLI utilities
- Train your own model variation on a similar dataset
- Test the imitation learning algorithm in manipulator control tasks
Example
lerobot-rollout \
--strategy.type=base \
--robot.type=unknown \
--robot.port= \
--robot.cameras="{ : {type: opencv, index_or_path: , width: 640, height: 480, fps: 30}}" \
--policy.path=ravediamond/lewam_jepa_pusht \
--task="Push the T-shaped block onto the T-shaped target." \
--duration=60Read more
DetailsTraining steps: 46163
Batch size: 64
Learning rate: 0.0001
Dataset: lerobot/pusht (206 episodes, 25650 frames)










Comments