Skip to main content

Bolt Runtime

Bolt any policy onto any robot.

A native C++ runtime that deploys learned robot policies in MuJoCo and on hardware, with one policy loop and plug-in observations.

boltrt --task tracking \
  --robot g1 --backend mujoco \
  --policy mimic-lite/v1_1/policy.yaml \
  --motion motions/walk1_subject1.npz
Modular observationsEach policy composes registered termsMimicLiteSONICyoursjoint_posgravityref_joint_posref_joint_velref_root_posref_root_oriref_body_posSeamless sim-to-realSim and real share one interfacepolicy loopRobotBackendread_state · submitMuJoCosimmujocoG1realg1RP1realrp1Yoursreal...Fast native C++Full policy step, median3.25×faster than the Python stackPython1.46 msBolt Runtime0.45 msIntel Core i9-13900H · 1 CPU threadLearn more →Learn more →Learn more →

Runs out of the box

MimicLite · BFM-Zero · GRIT · HEFT · HoloMotion · Humanoid-GPT · ScaleBFM · SONIC · TeleopIT · TWIST2

Supported policies and parity results →

Bring your own policy with a coding agent

The repository ships a boltrt-extend skill. Ask Claude Code or Codex to “deploy my checkpoint on G1” and it follows the same rules and checks as these docs.

About the skill →