In 2025, six students graduated from our laboratory: Five master students and one undergraduate student. Below is the title and a quick summary of each of their theses. If you are interested in knowing more, feel free to reach out!
Master students:
Fang "Michael" Zikun: "Optimizing Shelter Location in Disaster Evacuation Simulation Considering Elevation and Exhaustion"
In this thesis, Michael designs a model of energy and exhaustion to the agents of an earthquake evacuation simulator. This model allows him to represent how the difference in height (hills, climbs, flat roads) affect the movement of the agents during an evacuation in the simulator. He uses this model to explore the optimal location of earthquake shelters in the simulation of a hilly town.
Richard Alison: "A Study on Geographical Niching in the Co-Evolutionary Wolf Sheep Predation Simulation"
In this thesis, Richard studies a co-evolutionary predator-prey model, wherere both the predator and the prey can evolve the neural network that controls their movement. The thesis focuses on the introduction of height differences to the environment, and whether the assymetries introduced by this geography can help cause niching and/or speciation of the predator and prey agents. Additionally, Richard investigates what happens when the agents are given the ability to modify their environment by picking up and placing blocks (like endermen in Minecraft).
Ryota Kamimura: "Proposal of a Diversity Maintenance Method Using Niching in the Vehicle Routing Problem"
In this thesis, Kamimura studies how to improve the results of the (Mouti-Route) Vehicle Routing Problem. First he designs a method using the convex-hull algorithm that generate an initial set of approximate solutions that can be used as a starting point for any kind of solver. Second he develops an Evolutionary Algorithm that employs MAP-Elites to increase the diversity of the population by separating solutions with different geometric characteristics. The combination of these two techniques allows his method to achieve better solutions in standard VRP benchmarks.
Takahiro Suzuki: "Feature-Based Multi-Heuristic Search in Dynamic Environments for a Baba Is You Solver"
In this thesis, Suzuki develops a solver for the "Baba Is You" computer game. Baba is You is a complex puzzle game where the player is able to change the rules of the game during gameplay. Although it has a slight similarity to Sokoban, the search space is many times higher. Suzuki approaches this problem by designing a solver that groups and organizes the search space using features defined by different rulesets. This allows a divide-and-conquer approach that can solve much harder instances (levels) of the game, as tested against the game levels published in the "Keke is You" competition.
Yuuta Kobayashi: "Niching Differential Evolution with Sustained Subpopulation Diversity in the Search Space"
In this thesis, Kobayashi investigates the problem of Niching Optimization, where an optimizer tries to find several different optima in a multi-modal problem. First, he develops a new measure for estimating the diversity of a set of solutions that addresses a number of theoretical problems in current diversity measures. Then, he develops a niching version of differential evolution where subsets of the population are defined by their distance to each other and their ability of tracking separate optima in the problem. He evaluates his result in several niching benchmarks.
(Note: Kobayashi presented two papers related to this work at IEEE-WCCI 2024)
Undergraduate students:
Masato Notsu: "A studing in the movement patterns of agents in a Earthquake/Tsunami evacuation simulator using GIS data"
In his graduation project, Notsu re-designs an evacuation evacuation simulator created by Matsushima (a former student of mine). Matsushima's simulator was used as the basis for several evacuation simulation studies of later students, but was limited in that agents could only move on a well defined road graph. Notsu redesigns the simulation using a raster-based terrain model based on GIS data such as land use and height, allowing the agents to move freely anywhere on the map. This required a re-thinking of the pathfinding algorithm, which was the focus of this work.