Adaptive Motion Planning for Autonomous Mass Excavation

Patrick Rowe
doctoral dissertation, tech. report CMU-RI-TR-99-09, Robotics Institute, Carnegie Mellon University, January, 1999


Download
  • Adobe portable document format (pdf) (1MB)
Copyright notice: This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.

Abstract
Autonomous excavation has attracted interest because of the potential for increased productivity and lower labor costs. This research concerns the problem of automating a hydraulic excavator for mass excavation, where tons of earth are excavated and loaded into trucks. This application is commonly found in many construction and mining scenarios. In such applications, fast opera-tional speed of these machines is desired, because it directly translates to increased productivity.

A hydraulic excavator can be considered a large, four degree-of-freedom manipulator mounted on a tracked base. The bucket at the end of the manipulator is used for digging and depositing the excavated material into the trucks. A core technology required for automation is the motion planning of the excavator? manipulator. This research focuses on planning the free motion of the excavator, which begins after digging a bucket of material, and ends after the material has been deposited and the bucket has been returned to the digging area. The goal is to plan the excavator? motions such that autonomous task performance approaches that of a highly skilled human expert operator working in similar conditions.

Much of the prior research in autonomous excavation has focused on digging and related topics such as soil modeling and bucket-soil force interactions. Only a few researchers have looked into the free motion planning problem within the context of the mass excavation task. Also, much of the autonomous excavation research has concentrated on functionality, where simply digging a full bucket of material is good enough. In contrast, this research has explicit performance goals, placing importance on high productivity. Finally, while other work has been done in the difficult area of controlling large hydraulic machines, not much emphasis has been placed on the motion planning phase.

There are several characteristics about this problem that led to the motion planning approach that is presented in this thesis. The excavator? motions for each bucket loading cycle 1 are highly repetitive and deliberate, almost to the point of being scripted. However, the precise dig, dump, and truck locations do change from cycle to cycle. The hydraulic actuation system of the excavator is power-limited and highly non-linear, making it difficult to model. The operation proceeds quickly, with many buckets being dug and loaded in a short amount of time.

With these characteristics in mind, we have developed a motion planning approach known as parameterized scripting. A script describes the task as a series of simple steps. The parameters of the script define both the specific goals for each script step and the transitions between steps. Different script parameter values are computed for each bucket loading cycle based on the current task conditions. The parameter values affect both the operational speed and the accuracy in achieving desired task goals.

Because of the modeling difficulties, the script parameter values are computed using information about the excavator? own performance, which is gathered on-line during task execution. The excavator? performance, resulting from a particular parameter set, is evaluated and stored in a data base. Memory-based learning techniques are used to generalize across parameter sets and find the best set of parameter values for the given task conditions. The vehicle motion itself is also analyzed to help in the search for the best parameter values and rapidly improve task performance.

The adaptive motion planning system has resulted in autonomous performance that approaches a skilled operator in the short term and outperforms him in the long term under our testing conditions. The autonomous excavator? motions are also more accurate and consistent than a human?. The adaptive motion planning approach provides a highly flexible system. Because the adaptive motion planning system uses data gathered on-line, it can be used on any excavator with any con-trol system in any worksite conditions. The excavator can modify its behavior to achieve maximum productivity in its current working environment.

1. A loading cycle consists of the excavator? manipulator moving the bucket to the truck, dumping the material, and moving back to the dig area.


Notes

Text Reference
Patrick Rowe, "Adaptive Motion Planning for Autonomous Mass Excavation," doctoral dissertation, tech. report CMU-RI-TR-99-09, Robotics Institute, Carnegie Mellon University, January, 1999

BibTeX Reference
@phdthesis{Rowe_1999_535,
   author = "Patrick Rowe",
   title = "Adaptive Motion Planning for Autonomous Mass Excavation",
   booktitle = "",
   school = "Robotics Institute, Carnegie Mellon University",
   month = "January",
   year = "1999",
   number= "CMU-RI-TR-99-09",
   address= "Pittsburgh, PA",
}