facebookresearch/habitat-lab

Features

• Embodied AI Platforms - Provides an integrated platform for training and evaluating embodied AI agents in photorealistic 3D indoor environments.
• Reinforcement Learning Integrations - Provides pre-built PPO training loops with vectorized environments and reward shaping for embodied AI tasks.
• Reinforcement Learning Training - Trains agents on embodied AI tasks using PPO-based reinforcement learning with provided baselines.
• Embodied Agents - Trains agents to perform navigation and manipulation tasks inside simulated indoor environments.
• Reinforcement Learning Frameworks - Provides PPO-based training pipelines and benchmark tasks for developing navigation and manipulation agents.
• Simulation Environment Libraries - Provides access to core simulation components for environments, tasks, datasets, and simulators to benchmark agent performance.
• Interactive 3D Simulators - A high-performance simulator that renders realistic indoor scenes and supports physics-based interaction for embodied AI research.
• Modular Sensor-Actuator Frameworks - Organizes agent perception and action into modular sensor and actuator classes that can be mixed and matched per task.
• Embodied AI Task Loaders - Loads embodied AI tasks from structured dataset files defining scene configurations, object placements, and success criteria.
• Registry-Based Component Loaders - Separates task logic from environment simulation using interchangeable configuration files and registry-based component loading.
• Configuration-Driven Pipelines - Builds simulation and training pipelines by composing YAML configuration files that specify sensors, actuators, and task parameters.
• ROS Simulation Bridges - Connects the 3D simulator to external robotics frameworks through a ROS interface for real-time control and sensor feedback.
• ROS Simulation Bridges - Bridges simulation with the Robot Operating System for physics-based robotics development and asset access.
• Agent Behavior Visualizers - Renders top-down maps, agent trajectories, and 3D scene views using off-screen rendering and matplotlib integration.
• Robotics System Integrations - Connects Habitat simulation to ROS for physics-based robotics development and real-time asset access.
• Habitat-ROS Bridges - Provides a bridge connecting Habitat simulation to the Robot Operating System for robotics development and testing.
• Agent Behavior Visualizers - Explores agent behavior through top-down maps, 3D environment views, and step-by-step task execution in notebooks.
• Interactive AI Demos - Runs interactive tutorials and notebooks to demonstrate agent behavior and explore 3D environments.
• Keyboard-Controlled Agent Demos - Ships interactive keyboard-controlled agent demos for manual testing and exploration of 3D environments.
• Rearrangement - Trains home assistants to rearrange objects within their habitat.
• Sim to Real Transfer - Provides infrastructure for training and deploying embodied agents.

Star history