model_act_test_gazeuse_resnet_15K_18_11
a LeRobot robotic arm policy
2 min readupdated
This model uses the ACT algorithm (Action Chunking with Transformers, an imitation learning method that predicts short sequences of actions rather than single steps) to control an so_follower robotic arm. The policy takes joint states and feeds from three cameras (front, side, and top) as input, outputting control actions. Trained on teleoperation demonstrations, it handles picking up a yellow marker and placing it into a pink box. It is built for researchers and practitioners working on home and lab robotics with the LeRobot framework.
Key features
- Supports three cameras (front, side, top) at 480x640 resolution
- Predicts 6-step control action sequences (action chunking)
- Compatible with the so_follower robot hardware configuration
- Trained on network demonstrations recorded at 30 FPS
What you can do
- Deploy the pre-trained policy on a physical so_follower robot for object sorting tasks
- Integrate the model into your own imitation learning experiments within the LeRobot ecosystem
- Test policy inference using LeRobot command-line tools
Example
lerobot-rollout \
--strategy.type=base \
--robot.type=so_follower \
--robot.port= \
--robot.cameras="{ : {type: opencv, index_or_path: , width: 640, height: 480, fps: 30}, : {type: opencv, index_or_path: , width: 640, height: 480, fps: 30}}" \
--policy.path=Tridex/model_act_test_gazeuse_resnet_15K_18_11 \
--task="take the yellow fluo and drop it inside the pink box" \
--duration=60Read more
DetailsLicense: apache-2.0
Dataset: 10 episodes, 4584 frames
Training parameters: 15000 steps, batch size 8, adamw optimizer, learning rate 1e-05










Comments