Sim2Real Deployment

Deploy Teleopit to control a physical Unitree G1 robot via g1_bridge_sdk (C++ DDS bridge).

tip

For Pico 4 / Pico 4 Ultra VR deployment, see the complete Pico VR Tutorial.

Input Sources

Source	Config	Documentation
Pico 4 / Pico 4 Ultra	--config-name pico4_sim2real	Pico VR
Offline BVH files	Default config	This page

Prerequisites

Hardware:

• Unitree G1 (29 DOF)
• Unitree wireless remote controller
• Network connection between robot and control PC
• Pico 4 headset, or offline BVH motion files

Software:

git submodule update --init --recursive
bash scripts/setup/setup_g1_bridge.sh
pip install -e '.[sim2real]'

For Pico path, also run: bash scripts/setup/setup_pico4.sh

Offline BVH Playback

python scripts/run/run_sim2real.py \
    controller.policy_path=track.onnx \
    real_robot.network_interface=eth0 \
    input.bvh_file=data/sample_bvh/aiming1_subject1.bvh

Remote Controller Mapping

Button	Action
Start	Enter STANDING
Y	Enter playback / MOCAP
A	Pause / Resume
B	Replay from start
X	Return to STANDING
L1+R1	Emergency stop (DAMPING)

Control Modes

Mode	Data Flow	Use Case
STANDING	Default pose -> RL policy -> joints	Startup, recovery, waiting
MOCAP	Pico/BVH -> retarget -> RL policy -> joints	Teleoperation / playback
DAMPING	Send damping command	Emergency stop

State Machine

                     +-----------------------------+
                     |    L1+R1 E-stop (any state) |
                     v                             |
  [IDLE] --Start--> [STANDING] --Y--> [MOCAP] --X--> [STANDING]
                           ^                       |
                           +----------Y------------+
    ^                                                  |
    +------------------Start---------------------------+
                           [DAMPING]

Pico MOCAP Sub-states

When using input.provider=pico4, the MOCAP mode has additional sub-states:

• ACTIVE: Normal live mocap tracking
• PAUSED: Freeze reference pose; robot maintains balance but stops following
• RESUMING: Clear realtime reference buffer, rebuild yaw/pivot alignment, smooth transition back to live mocap

Common Parameters

# Adjust control frequency
policy_hz=30

# Specify BVH file
input.bvh_file=data/sample_bvh/aiming1_subject1.bvh

# Pico timeout
input.pico4_timeout=30

# Pause/resume transition
pause_resume_transition_duration=1.5
pause_resume_warmup_steps=3

# Network interface
real_robot.network_interface=enp3s0

Standalone Standing Test

A minimal script for quick hardware and policy verification, independent of the main framework:

python scripts/run/standalone_standing.py \
    --policy track.onnx \
    --network-interface eth0

Supports --dry-run for safe timing benchmarks without sending motor commands.