Strict Subgoal Execution

Jaebak Hwang1, Sang-hyeon Lee1, Sang-jun Bae1, Yong-hyeon Jo1, Yi-sak Park1, Seungyul Han1
1AI Graduate School, UNIST
📄 Paper 🎓 arXiv 💻 Code (GitHub)
Abstract

In certain complex environments, learning tends to be unstable, so to mitigate this issue, we propose Strict Subgoal Execution (SSE). SSE is a graph-based Hierarchical Reinforcement Learning framework. Currently, there are few studies addressing this specific challenge, but SSE provides a stable approach by combining Goal-Conditioned Reinforcement Learning with structured subgoal planning.

Citation
@inproceedings{hwang2026sse,
  title={Strict Subgoal Execution},
  author={Hwang, Jaebak and Lee, Sang-hyeon and Bae, Sang-jun and Jo, Yong-hyeon and Park, Yi-sak and Han, Seungyul},
  booktitle={The Fourteenth International Conference on Learning Representations},
  year={2026},
  url={https://openreview.net/forum?id=cMpOvMuyYa}
}