Development · GitHub repo
Genesis World
Unified multi-physics simulation platform for physical AI and robotics development.
Genesis World is an open-source physics engine and simulation platform tailored for robotics and physical AI research. It integrates rigid bodies, FEM, MPM, SPH, and cloth solvers into a single unified scene with cross-platform GPU compilation and photorealistic rendering. The platform lets researchers simulate complex physical interactions and train reinforcement learning agents from a clean Python interface.
What it's for
- Train reinforcement learning policies for robotic arms, quadrupeds, and drones in batched parallel environments
- Simulate multi-physics interactions combining rigid bodies, fluids, granular materials, and deformable objects
- Render synthetic sensor data including depth cameras, LiDAR, and tactile sensors for embodied AI pipelines
- Perform batched inverse kinematics and trajectory planning at scale on GPUs
Features
- Unified multi-physics engine supporting Rigid, FEM, MPM, SPH, PBD, and IPC solvers in one state
- Cross-platform Quadrants compiler lowering Python kernel code to CUDA, ROCm, Apple Metal, Vulkan, and CPU
- Pluggable photorealistic rendering backends including Nyx, Luisa ray tracing, and Pyrender
- Native support for common robotics asset formats (URDF, MJCF, OBJ, GLB, USD)
- Built-in sensor simulation for depth cameras, IMU, LiDAR, tactile arrays, and contact forces
- Fast batched and heterogeneous simulation for high-throughput robot learning
Pricing
Open source. Free and open-source under the Apache-2.0 license.
Details
- Type
- GitHub repo
- Category
- Development
- Topic
- Robotics Simulation
- Open source
- Yes
- Platforms
- Linux, macOS, Windows, Code library
- For
- Robotics researchers, machine learning engineers, and embodied AI developers.
- Alternative to
- NVIDIA Isaac Gym, MuJoCo, PyBullet
- Tags
- robotics, simulation, physics engine, embodied ai, reinforcement learning, python
Why it was saved
- @evolving.ai: Simulates complex physical phenomena (rigid bodies, soft robotics, fluids, non-convex collisions) up to 10–80x faster than NVIDIA Isaac Gym, with cross-platform GPU support and text-to-4D scene/skill generation.
Related
Open-source collection of runnable AI agent, skill, and RAG templates
Open-source Python library and API for prompt-based web scraping using LLMs
Lightweight open-source Python library for building code-writing and vision AI agents
Python interactive CLI toolkit for automated OSINT, network reconnaissance, and security checks
Architecture pattern for securely running code-executing agents inside isolated micro-VMs
Tooling and authentication layer connecting AI agents to external applications
Updated