Research

My research has focused on multi-agent reinforcement learning and human–AI coordination, done at Oxford's Foerster Lab for AI Research.

• ICML 2025spotlight

  Ad-Hoc Human-AI Coordination Challenge

  Tin Dizdarević, Ravi Hammond, Tobias Gessler, et al., Jakob Foerster

  A Hanabi-based benchmark for coordinating with unseen human partners, using human proxy agents trained on a large human dataset as cheap, reproducible evaluation partners.

  CodeFLAIROx/ah2ac2

• ICLR 2025first author

  OvercookedV2: Rethinking Overcooked for Zero-Shot Coordination

  Tobias Gessler, Tin Dizdarević, Anisoara Calinescu, Benjamin Ellis, Andrei Lupu, Jakob Foerster

  Shows that limited state coverage is the root cause of zero-shot coordination gaps in Overcooked, and introduces a harder version with partial observability and stochasticity that requires forming protocols at test time.

  CodePart of JaxMARL  PosterICLR 2025

Projects

• 
  sep 2026benchmark

  DrivingBench: Can Vision-Language Models Drive a Toyota Corolla?

  Aditya Ramabadran, Simon Mahns, Tobias Gessler

  The first benchmark where general-purpose vision-language models drive a real car: they see camera frames and command steering and speed around a low-speed cone course while the car keeps moving, so latency is part of the task.

  Sitedrivingbench.com

• 
  ICIP 20221st place

  IEEE Video and Image Processing Cup

  With Vinzenz Dewor, Luca Beetz and ChangGeng Drewes, supervised by Christian Riess

  Detecting AI-generated images. Our FAU team won the grand prize at ICIP 2022 in Bordeaux.