v0.1

MorphMimic v0.1

Pretrained policy: morphmimic_policy.zip holds one ONNX policy for the Unitree G1, AgiBot X2 Ultra and Booster T2. It comes from the unified recipe (--robots g1_amber x2 t2) at 60,000 iterations, and works with both goal interfaces: motion and human. Unzip it, check it with sha256sum -c SHA256SUMS, and pass the folder as --deploy (see the README's "Pretrained policy" section).

New in this release:

• Human interface: the policy follows a PICO headset's body tracking directly, with all 24 joints or five points (pelvis, wrists, ankles), and no retargeting.
• g1_amber: the G1 with its waist roll and pitch damping raised to 3.0, for robots with weak waist motors.
• Held actuator delay (0 to 20 ms, drawn once per episode) is now the training default, for robustness to latency on hardware.
• Onboard teleop: the PICO service, the streamer, the policy and the robot node all run on the G1's own computer (deploy/onboard/).
• Updated X2 Ultra and T2 models: AgiBot's joint gains, and Booster's gains with their motor model.

Status: the policy is provided as is, with no warranty and no support. We have tested it on a real G1 only. On the X2 Ultra and the T2 it has run only in the sim, and their sim2real tests are waiting for the robots. Start any robot hung or held, and keep the remote ready to damp it.