Odin Navigation Stack

This open-source navigation stack, built upon the proprietary Odin1 spatial sensing module, provides advanced robotic intelligence capabilities for ground platforms. It targets roboticists and researchers seeking high-precision 3D mapping, robust relocalization, and sophisticated navigation features like dynamic obstacle avoidance, semantic object detection, and vision-language integration. The stack aims to enable autonomous navigation and richer scene understanding for robotic applications.

How It Works

The system leverages the Odin1 module's high-performance spatial sensing, which delivers precise 3D mapping, relocalization, and rich sensory streams (RGB, depth, IMU, odometry, point clouds). On this foundation, the stack implements dynamic obstacle-aware navigation using reactive local and global planners, semantic navigation that allows object-based movement via natural language commands, and vision-language models for contextual scene description. Its architecture is modular and based on ROS1, facilitating integration and customization.

Quick Start & Requirements

• Primary Install/Run: git clone --depth 1 --recursive, followed by ROS workspace compilation (catkin_make) within a Mamba/Conda environment.
• Prerequisites: Ubuntu 20.04, ROS1 Noetic, Unitree Go2 robot platform (or similar), NVIDIA Jetson (Orin Nano) or x86 with GPU, Unitree Go2 SDK, Conda/Mamba, PyTorch (CPU or Jetson-specific wheel), LLaMA.cpp for VLM.
• Setup: Requires significant setup including driver modification, environment creation, dependency installation, and ROS workspace compilation. Specific steps for Odin1 driver modification, USB rules, and Conda environment setup are detailed.
• Links:
  • NeuPAN paper: [Implicitly referenced]
  • Docker Setup: docs/DOCKER_SETUP.md
  • Tuning Guide: docs/TUNING_GUIDE.md
  • Unitree Go2 SDK: [Official guide required]

Highlighted Details

• High-Accuracy SLAM and persistent relocalization are handled by the Odin1 module (not open-sourced).
• Dynamic obstacle-aware navigation features reactive local planners and global path planning, with recent improvements including heading-aware A* search and dynamic obstacle memory.
• Semantic navigation enables navigation to detected objects using natural language commands.
• Vision-Language Integration allows for natural language scene descriptions using multimodal AI.
• Recent updates (June 2026) include NeuPAN planner improvements and a stuck-escape strategy.

Maintenance & Community

The project shows recent development activity with updates in June 2026. Specific community links (Discord, Slack) or details on core maintainers/sponsorships are not provided in the README.

Licensing & Compatibility

The README does not explicitly state the software license for the Odin-Nav-Stack itself. The Odin1 module is noted as proprietary. Compatibility is primarily for ROS1 Noetic on Ubuntu 20.04, with specific hardware recommendations.

Limitations & Caveats

Computationally intensive modules like YOLO, VLM, and VLN may require deployment on a separate companion machine due to onboard compute limitations. Standard ROS Navigation and custom planners are marked as "TODO" and not recommended. The NeuPAN planner may exhibit issues with precise stopping near target points, potentially requiring parameter tuning. The Odin1 module's core SLAM and relocalization functionality is not open-sourced.