Database filtering strategies for experience-based motion planning

Telagi, Praval

Permalink

https://hdl.handle.net/2142/129200 Copy

Description

Title

Database filtering strategies for experience-based motion planning

Author(s)

Telagi, Praval

Issue Date

2025-04-15

Director of Research (if dissertation) or Advisor (if thesis)

Amato, Nancy M

Department of Study

Siebel School Comp & Data Sci

Discipline

Computer Science

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

Motion Planning, Machine Learning

Abstract

Learning from previous experience in motion planning has been shown to significantly reduce search time in new, unseen motion planning problems. Often, these experiences are stored in a database as paths and a new problem will query the database for the most relevant solutions. Prior work in experience based motion planning tends to use limited information, such as only task information, about the motion planning problem when querying a database for solutions. Additionally, many prior methods fail to provide a principled way of generating a good experience database that allows their algorithm to perform optimally. We study the effects of a variety of filtering methods on a database for a novel motion planning algorithm Path Database Guidance (PDG) to understand which method consistently minimizes the average number of collision checks performed during search. Specifically, we analyze augmenting the database by performing random filtering, cluster-based filtering, and model-based filtering. Our results illustrate the significant impact that databases can have on the performance of PDG in solving various motion planning problems. Lastly, we show our model-based filtering method consistently minimizes the average number of collision checks across problem domains using information about the environment.

Graduation Semester

2025-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129200

Copyright and License Information

Copyright 2025 Praval Telagi

Owning Collections