Genesis-Embodied-AI/Genesis
Genesis
Genesis is an embodied AI simulation platform and parallelized robotics simulator designed for training general-purpose robotic agents. It integrates a physics engine for robotics that calculates collisions and movements for rigid bodies, soft tissues, and fluids, alongside a photorealistic 3D rendering engine.
The platform features a domain randomization framework to vary environment parameters across parallel simulations, aiding in sim-to-real transfer. It supports the integration of real-world captured light fields and Gaussian splatting to provide photorealistic backgrounds within simulated scenes.
The system covers multi-modal sensor data collection for depth, lidar, and inertial units, as well as robotic agent control for managing joint states and kinematics. It further provides capabilities for synthetic sensor data generation and multi-environment parallel rendering to accelerate the collection of training data.
Features
- Embodied AI Platforms - Provides a comprehensive platform for training robotic agents in simulated environments before physical deployment.
- Parallelized Robotics Simulations - Runs many simultaneous environment instances on compute backends to accelerate data collection for machine learning.
- Sim to Real Transfer - Bridges the gap between simulation and reality using domain randomization and photorealistic rendering.
- Domain Randomizations - Implements a domain randomization framework to vary environment parameters and improve the robustness of trained robotic agents.
- Sensor Data Simulation - Generates synthetic multi-modal sensor data, including depth, lidar, inertial units, and contact forces.
- Robotics Simulators - Simulates robot movements and sensor interactions using physics solvers for rigid bodies, soft tissues, and fluids.
- Parallelized - Distributes simulation loops across hardware threads to execute multiple environments simultaneously for faster learning.
- Physically Based Rendering - Implements physically based rendering and path-tracing to generate photorealistic frames of objects and terrain.
- Robotics and Control - Provides comprehensive tools for robotic system automation, including kinematics and joint state control.
- Parallel Scene Simulations - Executes multiple independent scene instances in parallel to generate diverse training data via domain randomization.
- Dynamics Solvers - Integrates multiple computational solvers to simultaneously simulate rigid bodies, soft tissues, and fluids.
- Gaussian Splatting - Integrates real-world captured light fields and Gaussian splatting to provide photorealistic backgrounds.
- Robot-Attached Sensor Simulations - Mounts simulated sensors to specific entities with offset transforms that update automatically during simulation.
- Parallel Loop Transformers - Automatically distributes top-level execution loops across available hardware threads to improve performance.
- Multi-Mode Rendering Strategies - Provides a rendering strategy that manages the distribution of tasks across multiple parallel environments for diverse training data.
- 3D Rendering Engines - Provides a rendering system for path-traced frames and integrated light field scenes for realistic sensor data.
- Gaussian Splatting - Integrates 3D Gaussian splatting to import real-world captured light fields as photorealistic backgrounds.
- GPU-Accelerated Compilers - Compiles high-level functions into optimized parallel kernels for various graphics and compute backends.
- GPU-Resident Memory Managers - Implements a GPU-resident memory strategy to minimize latency between physics calculations and rendering.
- Rendering Randomizers - Varies environment parameters across parallel simulations to improve the robustness of learned robotic models.
- Parallelizing Compilers - Translates high-level functions into optimized parallel kernels for high-performance hardware execution.
- Simulation and Benchmarking - Generative and universal physics engine for robotics.
- Simulation Frameworks - Generative world platform for general-purpose robotics.
- Physics Simulation - Generative physics platform with GPU-accelerated parallel simulation.
