Date of Award:

12-2022

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Computer Science

Committee Chair(s)

Mario Y. Harper

Committee

Mario Y. Harper

Committee

Shuhan Yuan

Committee

Steve Petruzza

Abstract

A heuristic is the simplified approximations that helps guide a planner in deducing the best way to move forward. Heuristics are valued in many modern AI algorithms and decision-making architectures due to their ability to drastically reduce computation time. Particularly in robotics, path planning heuristics are widely leveraged to aid in navigation and exploration. As the robotic platform explores and navigates, information about the world can and should be used to augment and update the heuristic to guide solutions. Complex heuristics that can account for environmental factors, robot capabilities, and desired actions provide optimal results with little wasted exploration, but are computationally expensive. This thesis demonstrates results of research into simplifying heuristics that maintains the performance improvements from complicated heuristics.

The research presented is validated on two complex robotic tasks: stealth planning and energy efficient planning. The stealth heuristic was created to inform a planner and allow a ground robot to navigate unknown environments in a less visible manner. Due to the highly uncertain nature of the world (where unknown observers exist) this heuristic implemented was instrumental to enabling the first high-uncertainty stealth planner. Heuristic guidance is further explored for use in energy efficient planning, where a machine learning approach is used to generate a heuristic measure. This thesis demonstrates effective learned heuristics that simplify convergence time and accounts for the complexities of environment. A reduction of 60% in required compute time for planning was found.

Checksum

cd3e5c1369e60456c43a8c136784ff83

Download

Share

COinS

Copyright for this work is retained by the student. If you have any questions regarding the inclusion of this work in the Digital Commons, please email us at .