wuyi2121/SCAN-Planner

What it solves

SCAN-Planner addresses the challenge of low-level local planning for quadruped robots during long-range navigation. It provides a robust foundation for robots to avoid collisions while following routes, enabling them to perform complex upper-level tasks like autonomous exploration and vision-language navigation across different environments, including multi-floor settings.

How it works

The system acts as a spatial collision-aware local planner that integrates with sensing inputs (such as LiDAR or depth cameras) to navigate obstacles. It can operate in three distinct navigation modes: interactive 2D goal setting, keypoint-based multi-floor navigation, and reference-path tracking with local obstacle avoidance. The framework is compatible with both CPU and GPU rendering for local sensing and is designed to work with platforms like the Unitree Go2.

Who it’s for

This project is primarily for robotics researchers and developers working on quadruped robot navigation, embodied intelligence, and autonomous exploration.

Highlights

• Versatile Navigation Modes: Supports interactive goals, keypoint-based multi-floor movement, and path tracking.
• Flexible Sensing: Compatible with both LiDAR (e.g., MID360) and depth cameras (e.g., RealSense D435).
• Hardware Ready: Provides CAD files for sensor layouts to ensure reproducibility.
• Cross-Platform Support: Includes a simulator for testing and community-contributed ROS2 support.