SO101 Sock Ball Handover ACT Dualcam 60eps
ACT policy trained on fbsh96/so101-sock-ball-handover-v1.
- Robot:
bi_so101_follower - Dataset: 60 good episodes, 84,717 frames, 30 FPS
- Inputs:
observation.state[12],observation.images.front_cam[3, 480, 640],observation.images.hand_cam[3, 480, 640] - Output:
action[12] - Training: MI300X, batch size 16, 20,000 steps
- Final logged loss: 0.065
- Checkpoint source:
checkpoints/020000/pretrained_model
- Downloads last month
- 5