facebookresearch/habitat-sim

Features

• Embodied AI Platforms - Provides a comprehensive platform for training and benchmarking embodied agents in simulated 3D environments.
• 3D Environment Rendering - Builds and manages complex indoor and outdoor scenes with physically based rendering and lighting control.
• Sensor-Body Attachments - Attaches sensor and control nodes to simulated rigid bodies to enable physical interaction with the environment.
• Physics Simulation - Manages global physical constants and the simulation engine to govern object interactions.
• Simulation Environments - Implements high-performance simulators for testing robotic agents within complex indoor and outdoor 3D environments.
• Robot-Attached Sensor Simulations - Links virtual cameras and depth sensors to agent bodies with dynamic offsets for synchronized observations.
• Robotic Sensor Simulation - Generates synthetic observations using customizable sensors to mimic real-world robotic inputs like lidar and depth cameras.
• Robotics Simulators - Implements virtual environments and physics engines to test robotic hardware and control logic.
• 3D Physics Engines - Includes a native C++ core for calculating real-time rigid body physics and collisions in 3D space.
• Navigation Meshes - Employs pre-computed navigation meshes to ensure collision-free traversal and realistic movement in 3D scenes.
• Navigation Mesh Queries - Utilizes pre-computed navigation meshes and spatial queries to enable efficient agent movement through 3D scenes.
• Rigid Body Physics Engines - Features a rigid-body dynamics engine to calculate physical forces and collisions for non-deformable 3D objects.
• Multi-Body Dynamics Simulators - Calculates physical interactions for 3D objects using a high-performance multi-body dynamics system.
• Physically Based Rendering - Uses PBR shaders and baked global illumination to simulate photorealistic light interaction with 3D materials.
• PBR Rendering Configuration - Adjusts shader parameters, including lighting, exposure, and tone mapping, to achieve high-fidelity photorealistic visuals.
• Synthetic Data Renderers - Utilizes assets with baked synthetic global illumination and high-fidelity rendering to generate imagery for AI training.
• Force and Impulse Applications - Exerts linear forces and torques on rigid objects to control movement and environmental effects.
• Runtime Action Extensions - Enables the definition and registration of new agent behaviors at runtime via custom functors.
• Behavioral Functor Definitions - Provides a mechanism to create custom movement and sensor control behaviors using functors.
• Static Scene Backdrops - Provides specifications for the architectural backdrop and collision meshes of simulation scenes.
• Synthetic Dataset Generators - Generates large-scale photorealistic RGB, depth, and semantic datasets for computer vision model training.
• Semantic Scene Annotations - Associates 3D meshes with hierarchical category labels and bounding boxes for scene querying.
• Object Rearrangement Simulations - Includes specialized simulation capabilities for training agents to move and place objects in indoor environments.
• Indoor Scene Libraries - Provides high-resolution 3D scans of residential and commercial environments for agent training.
• Synthetic Data Generation - Produces synchronized multimodal sensor observations for training computer vision and robotics models.
• Robot Model Importers - Allows the import of robot specifications from standard files to simulate diverse hardware types.
• 3D Scene Dataset Loaders - Loads generic 3D scene data from industry-standard datasets to create virtual training environments.
• 3D Scene Importers - Imports environment layouts and object configurations from external files to establish simulation spaces.
• Type-Safe Configuration Managers - Stores and retrieves nested settings using type-safe values to control simulation behavior.
• Articulated Object Simulation - Imports complex objects with jointed movement to manage physical properties like mass and inertia.
• NavMesh Constraints - Constrains agent movement to a navigation mesh to simulate realistic traversal and avoid collisions.
• Rigid Body Object Configuration - Configures visual and physical properties for non-articulated items, including mass and collision settings.
• Simulation Behavior Extensions - Provides a system for extending simulation behavior by registering custom functors into the simulation loop.
• 3D Object Transform Management - Adds, positions, and rotates individual 3D object instances using local coordinate frames.
• Asset Blueprints - Creates metadata templates for rigid and articulated objects to instantiate them in a 3D scene.
• Asset Loading Systems - Imports meshes, textures, and materials from standard formats to populate simulation environments.
• Camera View Controllers - Defines the observer's position and orientation in 3D space to render specific perspectives.
• Lighting Systems - Manages lighting configurations and illumination models to control the visual appearance of 3D scenes.
• Semantic Metadata Queries - Retrieves hierarchical semantic metadata, including category labels and bounding boxes, for scene elements.
• Scene Initialization - Defines simulation worlds by linking assets and specifying placement and properties via JSON metadata files.
• 3D Scene Environment Configurators - Sets global simulation parameters like gravity and timestep through configuration templates.
• Material and Shader Configuration - Manages rendering settings and lighting contributions via shaders and blending modes to determine the visual appearance of objects.
• JSON Scene Configurations - Defines 3D environment layouts and object properties through external JSON metadata files.
• Multimodal Sensor Captures - Generates synchronized RGB, depth, and semantic imagery from the agent's perspective for AI training.
• Scene Lighting Configuration - Defines global illumination using light sources and selects specific shading models for the simulation environment.
• Scene Lighting Management - Creates, positions, and modifies light sources and shaders to control the visual environment of the simulation.
• Scene Lighting Setup - Defines point and directional light sources with specific positions and intensities to control scene lighting.
• Stereo Vision Simulation - Simulates a stereo camera pair by attaching multiple sensors with distinct spatial offsets.
• Native C++ Applications - Implements the simulation and rendering pipeline in native C++ to achieve the high throughput required for AI training.
• Simulation robotique - High-performance 3D simulator for embodied AI research.
• Scene Understanding - High-performance simulator for embodied AI and scene navigation.
• Simulation Frameworks - Simulation platform for training home assistant robots.
• Simulation Platforms - High-performance simulator for photorealistic 3D environments.
• Simulation Engines - Simulation platform for embodied artificial intelligence.

Historique des stars