Library

eCommons

 

Multi-Agent Planning Under Uncertainty

Simple item page
dc.contributor.authorAsfora, Beatriz
dc.contributor.chairCampbell, Marken_US
dc.contributor.committeeMemberFerrari, Silviaen_US
dc.contributor.committeeMemberJung, Malteen_US
dc.date.accessioned2024-04-05T18:46:04Z
dc.date.available2024-04-05T18:46:04Z
dc.date.issued2023-08
dc.description141 pagesen_US
dc.description.abstractThe development of foundational theory and validated algorithms for human-robot teams operating in complex environments, capable of adapting as knowledge of the environment and tasks evolves over time, is a crucial area of research. As data becomes available, there is a need for evolving planning strategies to effectively utilize the information. In the specific context of search and rescue (SaR) scenarios, the high-level goals are to locate survivors, simulate rescue by having humans meet survivors, and minimize the risk to human team members. Ultimately, we seek to improve human-cooperation under uncertainty, as well as team performance and safety. This thesis explores probabilistic driven multi-agent planning approaches, motivated by the challenges posed by SaR missions.First, we investigate the problem of multi-robot non-adversarial search. Uncertainty is present in the victim’s true location as only probabilistic information is available a priori. In this context, we seek to find the optimal (or near optimal) path that maximizes the likelihood that our search team can intercept the target given a mission deadline. We prove this problem to be NP-hard, and present the first set of Mixed-Integer Linear Programming (MILP) models to encompass multiple searchers, arbitrary capture ranges, and false negatives simultaneously. The adoption of MILP as a planning paradigm allows to leverage the powerful techniques of modern solvers, yielding better computational performance and, as a consequence, longer planning horizons than the previous state-of-the-art. We build upon the proposed models to incorporate the concept of danger,estimated through a human-robot shared scene perception scheme, allowing for environment knowledge to evolve throughout the mission. The trade-off between risk vs reward is explored through conditional planning, based on the distinct agents’ tolerances to danger. We then consider other tasks beyond search in our mission, and ultimately even the planned routes and completion of tasks are modeled in a probabilistic manner. We introduce a novel problem formulation that incorporates probabilistic knowledge of task requirements, dependencies between tasks and their relative locations, heterogeneity of agents’ capabilities and an environment that might change as the agents interact with it. Performance assessment of possible mission plans is thus based on probabilistic predictions and tangible reward concepts for team forming, another important aspect of SaR missions.en_US
dc.identifier.doihttps://doi.org/10.7298/3tc8-xv15
dc.identifier.otherAsfora_cornellgrad_0058F_13819
dc.identifier.otherhttp://dissertations.umi.com/cornellgrad:13819
dc.identifier.urihttps://hdl.handle.net/1813/114565
dc.language.isoen
dc.rightsAttribution 4.0 International*
dc.rights.urihttps://creativecommons.org/licenses/by/4.0/*
dc.subjectmulti-agenten_US
dc.subjectoptimizationen_US
dc.subjectplanningen_US
dc.subjectprobabilityen_US
dc.subjectroboticsen_US
dc.subjectsearch and rescueen_US
dc.titleMulti-Agent Planning Under Uncertaintyen_US
dc.typedissertation or thesisen_US
dcterms.licensehttps://hdl.handle.net/1813/59810.2
thesis.degree.disciplineAerospace Engineering
thesis.degree.grantorCornell University
thesis.degree.levelDoctor of Philosophy
thesis.degree.namePh. D., Aerospace Engineering

Files

Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
Asfora_cornellgrad_0058F_13819.pdf
Size:
2.93 MB
Format:
Adobe Portable Document Format
Download

Collections

Cornell Theses and Dissertations