ManifoldTechLtd/Odin-Nav-Stack

What it solves

This project provides a comprehensive navigation and intelligence stack for robotic platforms (specifically the Unitree Go2) using the Odin1 spatial sensing module. It addresses the challenge of enabling robots to move autonomously in complex environments by integrating high-precision 3D mapping, dynamic obstacle avoidance, and the ability to understand and interact with the environment using natural language.

How it works

The stack is built on a modular ROS1 architecture that combines several key components:

• Sensing & Mapping: Uses the Odin1 module for 3D mapping and persistent relocalization via compact binary maps.
• Neural Planning: Employs the NeuPAN end-to-end neural planner for high-performance local navigation, featuring heading-aware A* search and dynamic obstacle memory.
• Semantic Understanding: Integrates YOLOv5 for object detection and Vision-Language Models (VLMs) like SmolVLM to generate natural language descriptions of scenes.
• Navigation Control: Combines global path planning with reactive local planners to allow for autonomous navigation, including "stuck-escape" strategies to bypass obstacles.

Who it’s for

Robotics developers and researchers working with the Unitree Go2 or similar ground platforms who want to implement autonomous navigation, semantic object targeting, and VLM-based scene understanding.

Highlights

• Dynamic Obstacle Avoidance: Includes reactive local planners and specialized strategies for escaping stuck positions.
• Semantic Navigation: Allows users to navigate to specific objects using spoken or typed natural language commands.
• VLM Integration: Capable of generating contextual scene descriptions using multimodal AI.
• High-Precision SLAM: Supports real-time mapping and long-term relocalization.
• Docker Deployment: Provides a one-command launch setup for NVIDIA Jetson devices.