fbsh96/rebot_smolvla_flipbread_44eps_20260425_6000steps

reBot SmolVLA Flip Bread 44 Episodes - 6000 Steps

SmolVLA checkpoint fine-tuned for seeed_b601_dm_follower on merged LeRobot flip-bread-to-pot demonstrations collected on 2026-04-25.

This model continues training from the 3000-step checkpoint:

fbsh96/rebot_smolvla_flipbread_44eps_20260425_3000steps

Schema

• observation.state: 7D
• observation.images.front: (3, 480, 640)
• observation.images.wrist: (3, 480, 640)
• action: 7D
• action chunk: 50 x 7

Joint/action order:

shoulder_pan.pos
shoulder_lift.pos
elbow_flex.pos
wrist_flex.pos
wrist_yaw.pos
wrist_roll.pos
gripper.pos

Training

Merged dataset:

phi-media-lab/rebot_flipbreadtopot_20260425_44eps
44 episodes, 18426 frames, 30 FPS

Training stages:

20 steps: smoke test
1000 steps: 10-episode overfit
3000 steps: 44-episode run
3000 additional steps: continued from 3000-step checkpoint

Effective total on the 44-episode run: 6000 steps.

Final continued-training loss: about 0.036.

Validation

The checkpoint reloads with SmolVLAPolicy.from_pretrained(...) and outputs (1, 50, 7) action chunks.

Observed L20 latency:

• first call: about 535 ms
• steady calls: about 149 ms/chunk

Example first actions from three merged-dataset frames:

[2.954437, -14.011684, -24.574974, 9.224384, -3.540721, -13.049671, -92.255066]
[-38.249802, -123.130707, -110.822876, 83.909637, -2.882618, -23.338799, -99.232513]
[2.840845, 0.345646, -4.611893, 11.636082, 1.457067, -3.227002, 5.661499]

Safety

This is a reBot-native overfit validation model, not a certified autonomous control policy. Before real actuator execution, use logging-only validation, clipping, rate limits, joint limits, and emergency stop handling.