Interactive Motion Planning for Multi-agent Systems with Physics-based and Behavior Constraints
Public Deposited- Last Modified
- March 19, 2019
- Creator
-
Best, Andrew
- Affiliation: College of Arts and Sciences, Department of Computer Science
- Abstract
- Man-made entities and humans rely on movement as an essential form of interaction with the world. Whether it is an autonomous vehicle navigating crowded roadways or a simulated pedestrian traversing a virtual world, each entity must compute safe, effective paths to achieve their goals. In addition, these entities, termed agents, are subject to unique physical and behavioral limitations within their environment. For example, vehicles have a finite physical turning radius and must obey behavioral constraints such as traffic signals and rules of the road. Effective motion planning algorithms for diverse agents must account for these physics-based and behavior constraints. In this dissertation, we present novel motion planning algorithms that account for constraints which physically limit the agent and impose behavioral limitations on the virtual agents. We describe representational approaches to capture specific physical constraints on the various agents and propose abstractions to model behavior constraints affecting them. We then describe algorithms to plan motions for agents who are subject to the modeled constraints. First, we describe a biomechanically accurate elliptical representation for virtual pedestrians; we also describe human-like movement constraints corresponding to shoulder-turning and side-stepping in dense environments. We detail a novel motion planning algorithm extending velocity obstacles to generate collisionfree paths for hundreds of elliptical agents at interactive rates. Next, we describe an algorithm to encode dynamics and traffic-like behavior constraints for autonomous vehicles in urban and highway environments. We describe a motion planning algorithm to generate safe, high-speed avoidance maneuvers using a novel optimization function and modified control obstacle formulation, and we also present a simulation framework to evaluate driving strategies. Next, we present an approach to incorporate high-level reasoning to model the motions and behaviors of virtual agents in terms of verbal interactions with other agents or avatars. Our approach leverages natural-language interaction to reduce uncertainty and generate effective plans. Finally, we describe an application of our techniques to simulate pedestrian behaviors for gathering simulated data about loading, unloading, and evacuating an aircraft.
- Date of publication
- December 2018
- Keyword
- DOI
- Resource type
- Rights statement
- In Copyright
- Advisor
- Barber, Daniel
- Manocha, Dinesh
- Lin, Ming
- Alterovitz, Ron
- Mayer-Patel, Ketan
- Degree
- Doctor of Philosophy
- Degree granting institution
- University of North Carolina at Chapel Hill Graduate School
- Graduation year
- 2018
- Language
- Date uploaded
- January 4, 2019
Relations
- Parents:
- In Collection:
This work has no parents.
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Best_unc_0153D_18249.pdf | January 4, 2019 | Public | Download |