v0.4.0

Highlights

• Improved Pico realtime control with pico-bridge 0.2.1, ARMS mode, and retargeter-preserving mode/pause resets.
• Added optional LinkerHand L6/O6 sim2real control, including Pico gripper input and low-latency L6 vr_hand_pose.
• Added manual Pico sim2real HDF5 recording and an interactive Pico motion recorder for training NPZ clips.
• Refined the training data path with minimal HDF5 shards, explicit precompute, rewind sampling, and updated tracking rewards.

Assets

• ModelScope assets updated for v0.4.0.
• act_40k policy published as checkpoints/track.onnx and checkpoints/track.pt.
• Canonical robot assets published as archives/robot_assets.tar.gz.
• ModelScope model tag: v0.4.0.

Teleopit v0.3.0

Realtime input and runtime consolidation release.

Highlights:

• Pico 4 now uses the in-process pico-bridge 0.2.0 receiver
• Optional host camera preview streaming back to Pico through pico-bridge
• Removed the old ZMQ and separate onboard Pico paths
• Added UDP BVH realtime input and online sim configuration
• Shared sim/sim2real reference-window buffering, realtime buffer config, pause/resume realignment, and velocity smoothing
• Expanded multi-viewer support with fixed RGB camera viewer
• Split sim2real reference processing and safety handling into focused modules
• Updated the G1 MuJoCo XML asset with the fixed RGB camera

Compatibility note:

• Teleopit v0.3.0 targets pico-bridge 0.2.0. Do not upgrade the Pico receiver to pico-bridge 0.2.1, because 0.2.1 changed interface semantics that this Teleopit release does not target.

Validation:

• python scripts/dev/check_large_tracked_files.py
• pytest tests/ -v: 228 passed, 1 skipped

Teleopit v0.2.0

Pico and sim2real runtime release.

Highlights:

• Pico 4 teleoperation through pico-bridge 0.2.0 on the Teleopit host
• Optional camera preview streaming back to the headset
• G1 Bridge SDK integration using the C++ DDS bridge
• Simplified offline playback with keyboard controls
• Pico sim2sim keyboard mode state machine
• Standalone standing controller and realtime mocap catch-up
• Released model upgraded to the 30k checkpoint

Teleopit v0.1.1

Maintenance and data pipeline release.

Highlights:

• Dataset shard-only refactor
• Adaptive bin sampling
• External asset management through ModelScope
• Repository slimming

Teleopit v0.1.0

Initial public release.

Highlights:

• General-Tracking-G1 whole-body tracking training
• ONNX sim2sim inference
• Pico 4 VR teleoperation
• Unitree G1 hardware deployment