Robot Policy Deployment

Collect demonstrations on a real SO-101 arm and turn them into deployable ACT and SmolVLA policies.

Goals

• Teleoperate a leader-follower SO-101 setup
• Record a personal LeRobot dataset with two cameras
• Train and evaluate ACT from scratch
• Fine-tune and deploy SmolVLA on the same demonstrations

SO-101 Policy Pipeline

• RPD01 — Teleoperation

  Connect, calibrate, and teleoperate the leader and follower arms.

• RPD02 — Data Collection

  Record synchronized robot and camera demonstrations.

• RPD03 — ACT

  Train ACT from scratch and deploy it on the follower arm.

• RPD04 — SmolVLA

  Fine-tune SmolVLA and compare it with ACT.

Return to the Physical AI overview.