MT Real2Sim Tutorial: 仿真-真机统一算法框架
Published:
Overview / 项目概览
MT Real2Sim Tutorial is an open-source, unified simulation ↔ real-robot algorithm framework built on Isaac Sim 5.1 and ROS1 Noetic. The simulation and physical deployment share the same ROS1 interfaces and algorithm code, requiring zero modifications for real-world deployment.
MT Real2Sim Tutorial 是一个开源的仿真-真机统一算法框架，基于 Isaac Sim 5.1 和 ROS1 Noetic 构建。仿真与真机使用相同的 ROS1 接口和算法代码，实现零修改部署。
This project was developed during my internship at Manifold Tech Ltd. (留形科技), where I served as the project lead.
Key Features / 主要特性
- Real2Sim / 真实到仿真: Import 3D-scanned real-world environments (from Manifold Tech hardware such as Q9000, Pocket 2 Pro, Odin 1) directly into Isaac Sim as simulation scenes.
- Sim2Real / 仿真到真机: Replace the simulation bridge with a real robot driver; all navigation algorithms and demo code remain unchanged.
- Docker-based / Docker 容器化: Dual-container orchestration (Isaac Sim + ROS Noetic) with one-click launch via
start.sh. - Full Sensor Suite / 全传感器支持: RTX LiDAR, Camera, and IMU simulation with configurable intrinsics/extrinsics.
- RL Locomotion / 强化学习运动控制: Unitree Go2 quadruped robot with RL-based locomotion policy.
- Autonomous Navigation / 自主导航: Terrain analysis, local planning, and path following algorithms (C++).
Architecture / 系统架构
The framework uses a TCP bridge to decouple the Isaac Sim simulation from the ROS1 algorithm stack, enabling seamless switching between simulation and real-robot deployment:
- Isaac Sim container: RL policy, sensor suite, TCP bridge server
- ROS Noetic container: sim_bridge, navigation algorithms (C++), demo scripts, RViz