dfl-rlab/dddmr_navigation

What it solves

It addresses the limitations of traditional 2D navigation stacks (like Nav2) by enabling mobile robots to map, localize, and autonomously navigate in complex 3D environments. Specifically, it solves challenges such as multi-layer floor mapping, navigating ramps, and handling obstacle avoidance in stereo structures and 3D point cloud maps.

How it works

The project provides a comprehensive navigation stack consisting of several specialized packages:

• Mapping: Uses dddmr_lego_loam for 3D mapping and refinement.
• Localization: Employs dddmr_mcl_3dl for Monte Carlo Localization (MCL) in 3D.
• Perception: Handles 3D perception and sensor data via dddmr_perception_3d.
• Planning: Utilizes a global planner (dddmr_global_planner) and a local planner (dddmr_local_planner) to calculate paths and avoid obstacles in 3D space.
• Movement: Managed by dddmr_p2p_move_base.

Additionally, it includes an automated CI/CD test suite to validate LiDAR configurations (angles, pitch, and transforms) to ensure point cloud accuracy and ground projection before deployment.

Who it’s for

Robot developers and beginners building custom mobile platforms (such as quadruped robots like the Unitree Go2) who need 3D navigation capabilities on cost-effective hardware like 16-line LiDARs, Jetson Orin Nano, or Intel NUCs.

Highlights

• True 3D Navigation: Supports multi-level mapping and ramp navigation where 2D frameworks fail.
• Automated Validation: CI/CD suite for verifying LiDAR mount geometries and spatial transforms.
• Hardware Agnostic: Designed to work with various sensors, including multiple depth cameras and different LiDAR models.
• Simulation Support: Integrated Gazebo models for Unitree-go2 to test 3D navigation in simulation.