Non-Line-of-Sight Perception Method for Autonomous Haul Trucks in Open-Pit Mines Based on 4D mmWave Radar and LiDAR...

Area of Science:

• Robotics and Autonomous Systems
• Sensor Fusion
• Computer Vision

Background:

• Open-pit mining presents significant perception challenges due to large equipment occlusion and sensor limitations.
• LiDAR suffers from blind spots, while 4D mmWave radar faces multipath interference in mining environments.

Purpose of the Study:

• To develop a robust sensor fusion framework for autonomous mining vehicles that overcomes occlusion and multipath interference.
• To enhance perception system safety and reliability in challenging open-pit mining conditions.

Main Methods:

• Proposed a Blind-Spot Complementary Fusion (BSCF) framework integrating 3D LiDAR and 4D mmWave radar using geometric constraints.
• Implemented multipath suppression, calibrated spatiotemporal alignment, and spatial consistency verification for sensor data.
• Introduced a Volume Recovery Rate (VRR) metric to quantify spatial evidence in occluded regions.

Main Results:

• The BSCF framework effectively suppressed multipath interference and improved cross-modal proximity by 16.5%.
• Achieved an overlap-region Root Mean Square Error (RMSE) of approximately 0.18 m.
• Provided target-existence risk cues with up to 15.6% VRR in completely occluded areas.

Conclusions:

• The proposed fusion framework significantly enhances perception robustness for autonomous transportation systems in open-pit mines.
• BSCF offers a viable solution for detecting hidden targets and improving operational safety in challenging mining environments.