Advancements in the 'Multi-Agent Exploration for Three-dimensional Space' Framework
Authors
Nielsen, Kasper Johan Holmgaard ; Stubkjær, Thomas Boel ; Biegel, Andreas
Term
4. term
Education
Publication year
2023
Submitted on
2023-06-16
Pages
11
Abstract
MAES3D (Multi-agent Exploration Simulator for 3D Space) er en åben kildekode-simulator bygget i Unity til at teste og sammenligne udforsknings- og dækningsalgoritmer—altså metoder der styrer, hvordan robotter systematisk undersøger ukendte 3D-områder og dækker dem fuldt ud. Denne afhandling udvider den tidligere version ved at rette mangler og tilføje funktioner: agenterne (simulerede robotter) får længere sensorrækkevidde, udforskede områder kan visualiseres, kortgeneratoren er mere konfigurerbar, og kamera samt brugergrænseflade er forbedret. Derudover tilpasses en ny udforskningsalgoritme, Dual-Stage Viewpoint Planner (DSVP), fra 2D med én robot til 3D med flere robotter. Vi sammenligner DSVP med to eksisterende metoder—Random Ballistic Walk (RBW) og Local Voronoi Decomposition (LVD). Forsøgene viser, at RBW klarer sig dårligere end både LVD og DSVP, og at DSVP er den klare vinder. En ydeevnetest viser desuden, at MAES3D kan simulere mindst 10 agenter samtidig.
MAES3D (Multi-agent Exploration Simulator for 3D Space) is an open-source simulator built in Unity for testing and comparing exploration and coverage algorithms—methods that guide robots to systematically explore unknown 3D spaces and cover them completely. This thesis extends the previous version by fixing limitations and adding features: agents (simulated robots) get a longer sensing range, explored areas can be visualized, the map generator is more configurable, and the camera and user interface are improved. It also adapts a new exploration algorithm, Dual-Stage Viewpoint Planner (DSVP), from 2D with a single robot to 3D with multiple robots. We compare DSVP with two existing methods—Random Ballistic Walk (RBW) and Local Voronoi Decomposition (LVD). Experiments show that RBW performs worse than both LVD and DSVP, with DSVP being the clear winner. A performance test further shows that MAES3D can simulate at least 10 agents at the same time.
[This summary has been rewritten with the help of AI based on the project's original abstract]
Documents