MIT-SPARK/Khronos

What it solves

Khronos addresses the challenge of performing simultaneous mapping and localization (SLAM) in dynamic environments where both short-term movements (like people walking) and long-term changes (like furniture being moved) occur. Traditional SLAM often struggles to distinguish between temporary dynamics and permanent changes to the environment.

How it works

Khronos uses a unified spatio-temporal approach to maintain a "4D map" of a scene. It performs metric-semantic SLAM to track the robot's position while simultaneously segmenting and tracking objects. It reasons about the scene state across time, allowing it to detect when an object has moved temporarily or when a permanent change has occurred (e.g., a chair being removed from a room) as the robot revisits areas and closes loops.

Who it’s for

This project is designed for robotics researchers and engineers working on autonomous navigation, scene reconstruction, and spatial AI in environments that change over time.

Highlights

• Spatio-Temporal Reasoning: Handles both short-term dynamics (human actions) and long-term environmental changes.
• Metric-Semantic Mapping: Combines geometric mapping with semantic object detection and segmentation.
• Open-Set Segmentation: Supports open-set semantic inference, allowing the system to recognize objects not seen during initial training.
• 4D Visualization: Includes a visualizer to scrub through the scene's state at different timestamps, showing how the robot's understanding of the environment evolves.