Download
Habitat-Sim is a high-performance 3D simulator for embodied AI research, designed to run photorealistic indoor environments at thousands of frames per second. It offers GPU-accelerated rendering and a flexible sensor suite—RGB, depth, semantic segmentation, and more—so agents can perceive and act in realistic scenes. The engine is written in C++ with Python bindings and integrates physics, navigation meshes, and shortest-path planners to support tasks like point-goal navigation, rearrangement, and interactive manipulation. It ships with connectors to popular datasets and scene formats, plus tools for dataset generation and scene replay. Determinism and reproducibility are first-class goals, which is critical for benchmarking agents and comparing algorithms. Thanks to its speed and modular design, Habitat-Sim is widely used to prototype embodied agents, train at scale, and evaluate in standardized environments with consistent metrics.
Features
- GPU-accelerated renderer with RGB, depth, and semantic sensors
- Physics, navmeshes, and planners for navigation and interaction
- Python API over a C++ core for flexibility and speed
- Dataset connectors and tools for scene generation and replay
- Deterministic simulation for reproducible benchmarking
- Scales to thousands of FPS for large-scale training and evaluation
Project Samples
