SENSEI: Semantic Exploration Guided by Foundation Models to Learn Versatile World Models

SENSEI distills foundation-model feedback into a reward signal that drives semantically meaningful exploration, letting agents learn versatile world models without task rewards.

How to run

• motif/ — annotate datasets with a VLM/GPT and train the Motif reward model (see motif/README.md).
• sensei_dreamer/ — train the SENSEI / DreamerV3 agents using the learned reward model (see sensei_dreamer/README.md).

Citation

@inproceedings{sancaktar2025sensei,
  title     = {{SENSEI}: Semantic Exploration Guided by Foundation Models to Learn Versatile World Models},
  author    = {Sancaktar, Cansu and Gumbsch, Christian and Zadaianchuk, Andrii and Kolev, Pavel and Martius, Georg},
  booktitle = {Proceedings of the 42nd International Conference on Machine Learning (ICML)},
  year      = {2025}
}