Autonomous Robotic Dozing for Rapid Material Removal

The autonomous control of vehicles interacting forcibly with their environment, such as dozers and tractors, is an unsolved and challenging problem. Forces and motions are inherently coupled between the tool and the means of vehicle propulsion. Furthermore, they are often operated within uncertain and unstructured environments, such as those encountered in underground mining. There is a growing industrial interest in the development of robotic vehicles to improve productivity, efficiency and safety in mining and construction. This paper focuses on the modeling and control of autonomous robotic dozing for a material removal operation. A dozing process model has been developed based on observations of both a fullscale dozer and a scaled-down version. The model characterizes the dynamic interactions between the blade position, material accumulation on the blade, material distribution in the environment, and the motions of the dozer. The dozing control objective is to remove the loose material as rapidly as possible by driving forwards at full power while automatically raising/lowering the blade in response to sensor measurements. Two different controllers were developed to meet this objective. The first controller is based on a set of heuristic rules, and the second is an optimal controller based on the dynamic model. An instrumented scaled-down robotic dozer and dozing environment, designed to emulate the full-scale operation, are used to implement the controllers and compare their performances over multiple dozing passes. Experimental results are presented showing that the model-based controller increased the material removal rate by 33% compared to the rule-based controller. Lastly, technologies for full-scale implementation are discussed, followed by proposed future work and conclusions.