In this thesis, you will explore how swarm behavior can emerge in teams of mobile robots through simple local interaction rules enhanced by Large Language Models (LLMs). Natural swarms such as ants, fish, or bird flocks exhibit robust collective behavior without centralized control. They coordinate through local interactions, adapt to dynamic environments, and collectively solve complex tasks such as exploration, aggregation, and dispersion. Translating such decentralized intelligence to robot swarms remains a key challenge in swarm robotics.

The goal of this thesis is to investigate how LLMs can be used to generate, adapt, or refine local interaction rules that produce useful emergent swarm behavior in teams of ground robots. Instead of explicitly programming global coordination strategies, the student will study how simple local rules—such as attraction, repulsion, alignment, signaling, or neighbor-aware movement—can be shaped through language-guided reasoning to produce collective swarm behaviors. These behaviors may include flocking, dispersion, clustering, cooperative exploration, or collective target tracking.

Possible tasks include:

• Designing decentralized local interaction rules for ground robot swarms
• Using LLMs to generate or adapt swarm rules from high-level behavioral goals
• Studying emergent collective behaviors such as flocking, aggregation, and dispersion
• Evaluating robustness under communication limits, sensing uncertainty, and robot failures
• Comparing LLM-generated swarm behaviors with hand-designed rule sets
• Testing in simulation and optionally on real ground robot platform

If you are interested in this topic, please send an email to both Prof. Groß and Ecem Isildar including a brief motivation letter, your CV, and your current transcript of records.

Core data