DynaFusion-SLAM: Multi-Sensor Fusion and Dynamic Optimization of Autonomous Navigation Algorithms for Pasture-Pushing...

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Existing autonomous navigation systems struggle with accuracy in complex outdoor environments like pastures.
• Limited research exists on multi-sensor fusion algorithms for autonomous navigation in challenging agricultural settings.
• Challenges include low fusion degrees and insufficient path accuracy for robotic operations.

Purpose of the Study:

• To propose a multimodal autonomous navigation system using a loosely coupled Cartographer-RTAB-Map architecture.
• To achieve high-precision mapping and robust path planning in complex pasture environments via laser-vision-inertial fusion.
• To enhance the cruising accuracy and operational efficiency of robots in outdoor settings.

Main Methods:

• Selected Cartographer as the front-end odometer due to its memory efficiency in large-scale scenarios.
• Implemented a two-way position optimization mechanism involving Cartographer for mileage estimation and RTAB-Map for fusing depth camera point clouds and laser data.
• Utilized Extended Kalman Filter (EKF) for fusing IMU and odometer data, Dijkstra's algorithm for global path planning, and Timed Elastic Band (TEB) for local path planning.

Main Results:

• The proposed system significantly improved mapping accuracy in a simulated pasture, reducing maximum absolute error from 24.908 cm to 4.456 cm.
• Multi-source odometry fusion effectively mitigated large-scale map offset and drift.
• Navigation accuracy tests showed reduced Root Mean Square Error (RMSE) by 26.7% and Standard Deviation (Std) by 22.8% compared to AMCL.

Conclusions:

• The multimodal autonomous navigation system demonstrates superior performance in complex simulated pasture environments.
• The fusion of multi-sensor data and optimized localization mechanisms leads to enhanced mapping and navigation precision.
• The system successfully enabled autonomous navigation for a robot, meeting task requirements for multi-point traversal and return.