Real-Time Dynamic Terrain Mapping for Autonomous Mining Operations

Autonomous mining machines must maintain an accurate model of the terrain they work in to support material movement planning and enable autonomous excavation. Terrain mapping is particularly challenging as these machines continuously modify the terrain they work in, often in the presence of other machines that simultaneously transform the same environment. This paper focuses on developing a capability to accurately create and maintain terrain maps that quickly reflect changes in the environment. The proposed approach extends the well-known height grid terrain representation to include a hidden Markov model per cell, allowing for confidence-based mapping of the constantly changing terrain that is robust to sensor noise. Hidden Markov models provide a probabilistic framework for handling uncertain observations and fusing sensor information to estimate the terrain height. The performance of the proposed mapping approach is evaluated on two field datasets: (i) an instrumented hydraulic excavator loading blasted ore into trucks, and (ii) a track-type tractor stacking material in a coal stockyard. The evaluation demonstrates the ability to accurately update the terrain estimate using the machine’s pose and point cloud data captured from machine-mounted LiDARs. The significance of this work is its capacity to provide real-time and accurate maps of changing terrain to support autonomous excavation.