Drag the pusher (red) into the T-block (blue) to shove it onto the green target.
Record episodes, download the z2r-teleop-1 dataset, then train a policy
(see TRAIN-LOOP.md). Drag the exported bc_policy.onnx onto the
scene — then Run policy to watch it drive, or Perturb block to knock the T
askew and watch it recover (honestly — it won't always). Arrow keys also drive the pusher.