High-Level Job Planning for Automated Earthmoving

Abstract

High-level job planning strategies were developed which enable pile transfer and area clearing jobs to be performed autonomously by a robotic wheel loader. A job is first planned on a 3D surface model of a worksite by positioning graphical tools representing areas and approach directions for scooping, dumping and clearing material. The ground model can be from a recently-acquired surface scan, allowing the job to be configured ad-hoc without the prior need of a global map. Algorithms interpret the high-level plan and, based on an updated ground model, generate commands which ideally guide the job to completion with no further human input. Lower-level plans such as driving points are also represented graphically, allowing a remote supervisor to stay in-the-loop by monitoring the intentions of the machine and modifying the plans if necessary. Fully automated jobs were demonstrated in an earthmoving simulation environment developed using Matlab. The algorithms and search parameters for finding clearing paths and filling locations which worked in the simulator were also found to correctly generate commands using ground models obtained from manually-performed area clearing and filling tests using snow and gravel. As proofs-of-concept, a snow clearing test and two pile transfer tests with gravel demonstrated semi-automated work cycles with a robotic loader, whereby driving and joint actuation were computer-controlled, with transitions between separate actions commanded manually. The snow clearing test demonstrated updated paths being generated based on the changing state of the worksite. The planning tools and algorithms were also extended to jobs including dump trucks and multiple loaders, and applied to a large-scale simulated hillside excavation. Additional simulations evaluated the proposed alternative High Point (HP) method for generating scooping commands, which orients the loader towards the highest point in the pile or slope section from an adjacent stage point. This was compared with a Zero Contour (ZC) method which selects perpendicular scooping approaches along the bottom contour of the slope. Various excavation jobs with truck loading showed that assuming the same bucket filling efficiency, the HP method offers the advantage of a higher excavation rate due to its more limited driving pattern. For the larger plateau excavation jobs, the workspace was subdivided by scanning with the smaller rectangular Scoop Area (SA). It was found that compared with the ZC method, the HP method tends to achieve its maximum excavation rate with SAs which are narrower and longer. Factors which increased the amount of material to excavate per area, including a higher plateau and more surrounding slope collapse, were found to generally result in smaller SAs achieving higher excavation rates.