sim-to-sim

Rex-Loco-Deploy: A Platform-Adapted Reinforcement-Learning Locomotion and Deployment Framework for a Servo-Actuated Quadruped. A full stack SpotMicro project with Isaac Lab reinforcement learning, STM32 and Nvidia Jetson Orin Nano

Deterministic C++17/Gymnasium robot-navigation simulator with PPO, native batching, MuJoCo sim-to-sim transfer, domain randomization, and Unreal integration.

Migrating a ROS 2 diff-drive rover from Gazebo to Isaac Sim, with a measurement harness that quantifies where the two physics engines diverge (determinism, RTF, drive-gain coupling, inertia approximations) instead of asserting they match.

Evidence-led Unitree G1 standing and locomotion research with mjlab, MuJoCo, and PPO

MuJoCo 部署_sim to sim

Lite3 四足机器人强化学习训练与 MuJoCo Sim-to-Sim 部署

MuJoCo simulation runtime and RL policy runner for H2 humanoid robot

RL for dexterous in-hand manipulation — 5 robot hands incl. passive/mimic joints · Isaac Lab · PPO · 100% success rate

Migrating a ROS 2 diff-drive rover from Gazebo Harmonic to Isaac Sim 6.0.1: baseline A/B measurement, USD importer transfer checks, and five deliberately-caused divergence signatures.

Archived MuJoCo model/optimization investigations and limited shared-contact sim-to-sim benchmark; local source/status updated October 2026.

PPO teacher-student whole-body controller for the Unitree G1 humanoid, trained in Isaac Lab and tested sim-to-sim on PhysX and MuJoCo: no falls in 66,000 test episodes without kicks. Built as the low level for a GR00T / π0.5 VLA (stages 3-5, planned).

Isaac Lab → MuJoCo transfer pipeline with ROS2 integration · 4 modes: MuJoCo · ROS2+RViz2 · ROS2+MuJoCo · ROS2+MuJoCo+RViz2

Humanoid locomotion, navigation, and language-guided manipulation in simulation: Isaac Lab/MuJoCo studies, sensor tests, and reproducible evaluation gates.

Fine-tuning OpenVLA on 53,294 Isaac Sim observations — contact improves, target selection degrades

Research portfolio for a Unitree G1 motion-learning pipeline: video recovery, retargeting, RL tracking, and cross-simulation validation.

Playwright-style testing for robots: one pytest test, 19 robots (Franka, UR, Kinova, KUKA, xArm, ALOHA, Unitree, Spot...), five physics engines (MuJoCo, PyBullet, Drake, Genesis, Isaac Sim). Auto-waiting actions, expect() assertions, traces, deterministic replay, codegen.

PPO navigation policies for a differential-drive robot, trained in 4,096 GPU-parallel MuJoCo Warp arenas (baseline vs domain-randomized), tested for robustness, exported to ONNX and raced as the opponent in an Unreal Engine 5.8 game.