Autonomous Loading of Fragmented Rock: Admittance Control for Robotic Digging

Thumbnail Image
Dobson, Andrew Arthur
digging , autonomous , mining , loading , admittance , LHD , control , robotic
An admittance-based controller for autonomously digging in fragmented rock piles was developed and tested by using a 1 t surface loader, and a 14 t Atlas Copco Load-Haul-Dump (LHD) vehicle. This admittance-based Autonomous Loading Controller (ALC) had superior consistency over both digging at constant velocity, and manual digging. The ALC was able to dig 39 % more material in 61 % less time than manual operators over 26 dig attempts by using a 14 t LHD in an underground mine, although it did require 68 % more work. The ALC achieves these performance improvements by controlling the mechanical admittance between the bucket and the rock pile. In contrast with nearly constant velocity digging observed in manual operators, controlling mechanical admittance allows the ALC to alter the velocity of the bucket in response to the dig reaction forces. Maintaining a preset admittance results in the ALC avoiding high and low force areas below the surface of the rock pile, which are indicative of obstacles and pockets respectively. The ALC is both novel and important because it was developed for use in fragmented rock piles by using industry-scale equipment, compared to existing methods that use small scale excavators in soil or gravel targets. The major contributions discussed in this thesis are the ALC, which opens a whole new area of digging research, a tuning process for the ALC, a comparative analysis of the ALC against manual and constant velocity digging, a preliminary study of how the rock pile characteristics a_ect loading, and methods for biasing the ALC towards successfully exiting the rock pile. The end result of these contributions is an autonomous loading method that loads considerably more material more consistently than a manual operator. The ALC was also remarkably robust to changes in dig platform and in multiple dig targets, which suggests that it is easier to tune and operate than behaviour, learning, or other heuristic based methods. The ALC still requires a human operator to select the rock pile entry point, but the ALC could be used with existing remote control methods to dramatically improve performance. Furthermore, the ALC is a vital step towards realizing the productivity and consistency advantages of fully automating the LHD cycle.
External DOI