A Multimodal 3D Perception-Based Autonomous Robotic Architecture for Underground Mining: Predictive Planning and Dynamic Risk-Aware Control

Abstract

Risk inspection and mitigation in underground mining pose critical challenges due to unstructured environments, limited visibility, and geomechanical instabilities. This paper presents an autonomous quadruped robotic architecture based on multimodal 3D perception fusion for safe navigation and early hazard detection in complex mining scenarios. The system integrates LiDAR, computer vision, and inertial data through probabilistic fusion to generate consistent three-dimensional maps using real-time SLAM. A risk-aware predictive planning framework is formulated as an optimization problem with dynamic weighting of critical zones and safe trajectory generation under kinematic and dynamic constraints. A robust dynamic controller ensures stability and adaptability over irregular terrain and external disturbances. The architecture is validated through high-fidelity ROS~2 simulations and laboratory experiments emulating underground gallery conditions. Results confirm real-time computational feasibility and demonstrate improved obstacle avoidance robustness and superior chassis stabilization compared to reactive navigation strategies. The proposed integration of multimodal perception, predictive planning, and risk-oriented dynamic control establishes a scalable framework for safe automation in underground mining.