Intelligent control of autonomous rock excavation: Theory and experimentation

Abstract

Earthmoving is a common activity at mines, construction sites, hazardous waste cleanup locations, and road works. Expensive and sophisticated machines such as front-end-loaders (FEL), backhoe loaders, LHD loaders and front shovels are used for these excavation tasks. Autonomous excavation control for these machines has gained considerable attention in order to remove human operators from hazardous environments, improve productivity and utilization, reduce machine abuse, as well as decrease machine operating costs. However, automatic control of excavation tasks for many sites that require digging in rock cannot be implemented using existing factory-based automation techniques. For example, control of bucket motions by simply partitioning the terrain into a set of volumes where each equals the bucket capacity often does not work. Planning in this way is possible only when digging in the materials such as loose soils where bucket motion resistance through the media can be predicted. Resistance predictions are impossible and/or infeasible to generate for excavation in the environments which consists mainly of irregular rigid objects such as rock piles with oversized particles, since no means exists to predetermine subsurface bucket/material interactions that are required to preplan the bucket trajectory. As a result, bucket actions must be determined through on-line decision making based on sensory feedback of the current excavation status in the unpredictable, unstructured and dynamic rock excavation environment. This research proposes a control method for autonomous rock excavation. The control architecture is designed following the behavior-based control concept. That is, the rock excavation control problem is solved by decomposition of the complicated task into a variety of simple elements that can be implemented by excavation behaviors. However, this control approach presents a new structure and operational paradigm that is developed based on, but different from the traditional behavior control method. Here, the behaviors are chosen using fuzzy excavation situation assessment with guidance of excavation task planning which embodies excavation heuristics and human strategies. Task plans are formulated using finite state machines which integrate neural networks for decision making. This organizational structure has the capability to include more excavation goals and to adapt to different environments via learning. Excavation behaviors are performed by primitive and machine executable actions or action sequences structured using finite state machines and simple action arbitration rules. The actions of human FEL operators were observed and analyzed to extract basic bucket actions and define rules of arbitration for different actions or action sequences under particular excavation environments. Fuzzy logic is applied to implement each excavation action where fuzzy rules represent the human experience and heuristics that are intrinsically linguistic, and bucket excavation motions are evaluated based on insufficient and inaccurate input sensory data. A variety of experiments were performed to test the ability of the proposed control algorithm. The laboratory-based experimental autonomous excavation system consists of a robotic arm, an excavation testbed, a force/torque sensor mounted between the robot arm wrist and the excavation bucket, and a control computer. Various rock piles to simulate realistic excavation environments and conditions were generated in the testbed. With these experiments, the control algorithm has demonstrated the ability to execute real-time automated loading cycles effectively and efficiently in complex excavation environments and under difficult digging conditions, through the use of the flexible excavation behaviors.