Robot Learning with UrbanVerse

This section provides comprehensive documentation for training navigation policies in UrbanVerse’s urban simulation environments. UrbanVerse supports both reinforcement learning and imitation learning paradigms, enabling you to train robots ranging from wheeled delivery platforms to humanoid robots in diverse, photorealistic city scenes.

The robot learning documentation covers:

• Simulation Paradigms: Understanding synchronous vs. asynchronous multi-environment simulation and how scene layouts and digital cousin variants are distributed across parallel training environments

• Robot Configuration: Configuring different robot embodiments (wheeled, legged, humanoid) and their action spaces, including COCO, Unitree Go2, Unitree G1, and more

• General Configurations: Understanding the EnvCfg configuration system that defines scenes, observations, actions, rewards, terminations, curriculum, and events

• Reinforcement Learning: Complete guide to training RL policies with UrbanVerse, including environment configuration, reward design, curriculum learning, and training workflows

• Imitation Learning: Complete guide to training Behavior Cloning policies, including data collection, dataset formats, training configuration, and evaluation

Note

We recommend system requirements with at least 32GB RAM and 16GB VRAM for UrbanVerse. For workflows with rendering enabled, additional VRAM may be required. For the full list of system requirements for Isaac Sim, please refer to the Isaac Sim system requirements.

Before training policies, ensure you have completed the Installation process and are familiar with Quickstart Guide workflows.

• Simulation Paradigms
  • 🎭 Synchronous Simulation
  • 👬👬 Asynchronous Simulation
  • 🔄 Comparison: Synchronous vs Asynchronous
• Robot Configuration
  • Supported Robots
    • Wheeled Robots
    • Quadruped Robots
    • Humanoid Robots
  • Dynamic Robot Loading Logic
  • Robot Spawn Initialization
  • Summary
• General Configurations
  • EnvCfg: The Environment Schema
  • Scene Configuration
    • Key Parameters
  • Observation Configuration
    • Fields
  • Action Configuration
  • Reward Configuration
  • Termination Configuration
  • Curriculum Configuration
  • Dynamic Initialization
  • Summary
• Reinforcement Learning in UrbanVerse
  • Documentation Overview
    • Working with Urban Scenes
    • Defining Robot Actions
    • What Your Policy Sees: Observations
    • Designing the Reward Function
    • When Episodes End: Termination Conditions
    • Progressive Learning: Curriculum Strategies
    • Environment Events and Initialization
  • The RL Environment Architecture
  • Basic Usage
  • Benefits of Reinforcement Learning
• Imitation Learning in UrbanVerse
  • Documentation Overview
    • Introduction to Imitation Learning
    • Collecting Expert Demonstrations
    • Demonstration Dataset Format
    • BC Training Configuration
    • UrbanVerse IL APIs
    • End-to-End Example
    • Evaluating BC Policies
    • Best Practices and Tips
  • Basic Usage
  • Benefits of Imitation Learning