On multi-agent hybrid system verification: Modeling in verse, star sets, and user studies

Braught, Katherine

Permalink

https://hdl.handle.net/2142/125597 Copy

Description

Title

On multi-agent hybrid system verification: Modeling in verse, star sets, and user studies

Author(s)

Braught, Katherine

Issue Date

2024-07-15

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

Mitra, Sayan

Department of Study

Siebel Computing &DataScience

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• verification
• reachability
• hybrid systems

Abstract

Hybrid systems have become the standard mathematical framework for modeling and verifying the safety of cyber-physical systems. Many effective tools for verifying hybrid systems exist; however, using these tools requires the user to write a model according to the tool’s specifications, which has a steep learning curve in the form of training in formal logics. The Verse library was developed to address this problem, allowing users to write their specifications in Python instead of using specialized logics. While over 200 engineering students have used Verse, its usability has not yet been formally evaluated. In this thesis, we present a preliminary user study of Verse and use its findings to improve the tool further. We designed an assignment and surveyed the experience of undergraduate engineering students who used the tool for design validation in a course on safe autonomy. We found that the students gave the verification aspect of Verse an average score of 54 on the Software Usability Scale, and the simulation aspect an average score of 61. These scores indicate the tool is usable but needs improvement. The most common complaints about Verse were related to its speed and documentation. Throughout the rest of this thesis, we provide updated documentation of the Verse tool. We also explore using new data structures in the reachability algorithms to increase speed. We present the star set data structure, which is an efficient and precise generalization of zonotopes, and demonstrate its potential to speed up Verse through experimental evaluations.

Graduation Semester

2024-08

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/125597

Copyright and License Information

Copyright 2024 Katherine Braught

Owning Collections