Robot Simulation Tutorial

This directory contains selection guidelines for the robot simulation platform, environment configuration tutorials, platform practice documents, and competition materials.

Entry Documentation

• Simulation Platform Selection and Directory Navigation - Comparison of the positioning and applicable tasks of platforms such as MuJoCo, Isaac Sim, ManiSkill, Habitat, GenieSim, Genesis, and AirSim.
• ManiSkill Environment Simulation Configuration - Tutorial on setting up the ManiSkill simulation environment.
• Deployment Guide for Isaac Sim, Isaac Lab, and GR00T - Unified instructions for local Windows, Linux workstations, cloud servers, Docker, micromamba/venv, Isaac Lab, and GR00T.
• Home Robot Environment Configuration - Configuration of the home robot simulation environment.
• SIM1 Soft Body Simulation and Data Generation - Teleoperation of dual-arm fabric manipulation, replay, diffusion trajectory generation, and filtering processes.
• Reproduction of UniLab + MotrixSim Heterogeneous Robot RL Training - Running the MotrixSim backend, PPO training, resource monitoring, and video playback of UniLab on a 6GB GPU.
• Introduction to Whole-Body Planning Data Generation with HumanoidMimicGen - Decomposing the methods proposed by NVIDIA and others for converting limited human teleoperation teachings into thousands of loco-manipulation data, covering skill annotation, whole-body IK, motion planning, G1 benchmark, and sim-and-real co-training.
• Introduction to PhysicsNeMo Physical AI Solver - Understanding the Physics-ML framework after the migration of NVIDIA PhysicsNeMo/Modulus, and distinguishing the boundaries between neural operator/PINN surrogate and robot interaction simulators.

Platform Special Topic

• ManiSkill Detailed Documentation - Detailed usage guide for the ManiSkill simulation platform.
• Isaac Sim Local and Cloud Configuration Tutorial - Configuration for local workstations, cloud servers, Docker, pip, and micromamba.
• Isaac Lab + GR00T Cloud Deployment Tutorial - Deployment process of historical versions of Isaac Lab and GR00T in Alibaba Cloud environment.
• Genesis Environment Configuration - Environment configuration for the Genesis simulation platform.
• Genesis Visualization and Rendering - Practices in visualization and rendering of Genesis.
• Complete Experience with Genesis World 1.0 - Comprehensive learning sections covering the official architecture, Blackwell/CUDA environment, Franka simulation, Nyx high-fidelity rendering, and indoor asset import capabilities.
• GenieSim Configuration - Environment configuration for GenieSim.
• GenieSim3 Configuration - Environment configuration for GenieSim3.
• UniLab + MotrixSim Heterogeneous Training - Practices in robot RL training with separated CPU simulation and GPU learner.
• Generation of Full-Body Planning Data for HumanoidMimicGen - Generation of full-body manipulation data for humanoid robots, linking MuJoCo/robosuite benchmarks with GR00T/policy learning.
• PhysicsNeMo Physical AI Solver - Extended tools for CFD, structure, weather, PDE surrogate, and Physics-ML; not a main focus of robot simulation.

Simulation Resources

• resource file - simulation-related images and resource files.
• simulation challenge - competition materials related to robot simulation.

Recommended Learning Path

1. First, read Simulation Platform Selection and Directory Guide to determine whether your task belongs to manipulation, navigation, system deployment, or cutting-edge platform research.
2. In the manipulation and reinforcement learning area: Learn ManiSkill Environment Simulation Configuration , and then access ManiSkill Detailed Documentation .
3. In the high-fidelity rendering and system integration area: Study Isaac Sim, Isaac Lab, and GR00T Deployment Guide , and then select the appropriate path based on your local Windows, Linux workstation, cloud server, or GR00T reproduction target.
4. In the household task and complex interaction area: Learn Household Robot Environment Configuration .
5. In the soft body manipulation area: Study SIM1 Soft Body Simulation and Data Generation .
6. In the reinforcement learning training infrastructure area: Learn UniLab + MotrixSim Heterogeneous Robot RL Training Reproduction , and understand how CPU physical simulation and GPU policy learning are separated.
7. In the humanoid full-body manipulation data area: Read HumanoidMimicGen Full-Body Planning Data Generation Guide , and understand how a small amount of telemanipulation teaching can be expanded into policy training data through skill constraints, full-body planning, and randomization.
8. In the physical AI extension area: Read PhysicsNeMo Physical AI Solver Guide , and learn about the differences between neural operators, PINN, and surrogate models and robot simulators.

Environmental Requirements

The requirements vary greatly across different platforms. It is recommended to follow the configuration documents. Generally, it is necessary to:

• Python 3.8+ or platform-specified version.
• CUDA and NVIDIA drivers for GPU simulation or high-quality rendering.
• Dependencies corresponding to platforms such as MuJoCo, Isaac Sim, ManiSkill, and Habitat.
• NVIDIA GPU; higher memory configuration is recommended for complex rendering and large-scale parallel simulations.