1. recreate an existing collective behaviour model that has been published fairly recently (last 5 years), and expand on that research; possible starting points (expand your research by searching for the title on http://scholar.google.com and following “cited by” and “related articles”):
  2. refactor the FRIsheeping source code to the ECS stack and use of computeShaders; you will be given access to the codebase of FRIsheeping including all assets; expand the environment, we aim for simulation of larger numbers of sheep (500-1000, anything more is a plus); the end goal should be also to develop a herding algorithm (possibly even collaborative) that is based purely on visual information and is better than the one in the current codebase (simplicity is the key). The topic can also be worked on by two groups, one working on the optimization of rendering, the other on the herding algorithm. 
  3. reimplement the current herding algorithm by means of Fuzzy Logic, attempt learning the behaviour by means of genetic algorithms
  4. use genetic algorithms, ml-agents, deep learning, or any other approach for learning the herding task; an extension of topic 1; the end goal is that the behaviour of the shepherd is not scripted, but learned; since learning takes time, refactoring of the FRIsheeping source code is a must
  5. implementation of a predator prey model for the study of fish schooling and hunting tactics in Unity; must be besed on the ECS stack (see also Boids in https://github.com/Unity-Technologies/EntityComponentSystemSamples) and should extend works from Demšar et al. (we will cover these in detail during the course)
  6. extension of Demšar et al. https://doi.org/10.1016/j.ecolmodel.2015.02.018, in view of Lambert et al. https://doi.org/10.1101/2021.05.25.445573, evolve choice of one of described targeting tactics.
  7. a topic by a group's own choice; this should have a clear foundation based on existing research - present a 10 sentence paragraph explaining the problem that is supported by at least two scientific references.

Related literature and tools:

  1. Escaping from Predators: An Integrative View of Escape Decisions, https://publicism.info/science/predators/
  2. Wu W. et al. 2021. Visual information based social force model for crowd evacuation, 10.26599/TST.2021.9010023
  3. Ishikawa Y. et al. 2021. Foids: bio-inspired fish simulation for generating synthetic datasets, 10.1145/3478513.3480520
  4. Wolter Jolles J. et al. 2021. Both Prey and Predator Features Determine Predation Risk and Survival of Schooling Prey, 10.1101/2021.12.13.472101
  5. Millington I. & Funge J. 2009. Artificial Intelligence for Games.
  6. Richmond P. et al. 2023. Fast Large-Scale Agent-based Simulations on NVIDIA GPUs with FLAME GPU.


마지막 수정됨: 월요일, 9 9월 2024, 9:29 AM