# A Subgoal-driven Framework for Improving Long-Horizon LLM Agents

The framework separates inference-time subgoal planning from MiRA, a reinforcement-learning method that supplies milestone rewards. WebArena-Lite experiments evaluate these mechanisms for long-horizon web navigation.

- Catalog URL: https://awesomecua.github.io/resources/arxiv-2603-19685/
- Original source: https://arxiv.org/abs/2603.19685
- Kind: paper
- Review: curated
- Scope: core
- Environments: web
- Topics: planning, reinforcement-learning, reward-models, long-horizon, evaluation
- Evidence checked: 2026-10-17T12:31:16Z
- ID: arxiv:2603.19685

## Limitations

Reviewed against the complete retained official abstract. Experimental findings and release availability have not been independently reproduced or verified.

- Record JSON: https://awesomecua.github.io/resources/arxiv-2603-19685/record.json

## Authors

Taiyi Wang, Sian Gooding, Florian Hartmann, Oriana Riva, Edward Grefenstette

## Citation exports

- BibTeX: https://awesomecua.github.io/resources/arxiv-2603-19685/citation.bib
- CSL JSON: https://awesomecua.github.io/resources/arxiv-2603-19685/citation.csl.json

## Recorded arXiv snapshot

- Version: 1
- Version source: https://arxiv.org/abs/2603.19685v1
- First submission: 2026-03-20T06:32:16Z
- Version updated: 2026-03-20T06:32:16Z
- Abstract SHA-256: 5538933a0e3d11df6ff4d767849979703d83412599fbbb0f8b432c4d9850633b
- Abstract hash normalization: sha256-utf8-whitespace-collapsed-v1

## License declarations

- code: not recorded
- data: not recorded
- document: not recorded
- weights: not recorded

## Primary evidence

- https://export.arxiv.org/api/query?id_list=2602.19870%2C2602.22769%2C2602.23166%2C2602.23320%2C2603.00198%2C2603.00842%2C2603.00991%2C2603.01050%2C2603.01145%2C2603.02236%2C2603.02795%2C2603.03278%2C2603.04549%2C2603.06713%2C2603.07106%2C2603.07670%2C2603.07853%2C2603.08706%2C2603.09716%2C2603.10002%2C2603.10664%2C2603.11139%2C2603.11320%2C2603.12483%2C2603.13026%2C2603.13247%2C2603.15714%2C2603.16065%2C2603.16289%2C2603.16839%2C2603.17216%2C2603.17378%2C2603.19313%2C2603.19685%2C2603.21489%2C2603.22527%2C2603.23638%2C2603.24157%2C2603.24639%2C2603.27646%2C2603.28069%2C2603.28407%2C2603.29418%2C2603.29847%2C2604.00528%2C2604.00886%2C2604.01194%2C2604.01438%2C2604.01560%2C2604.02985&max_results=50 (retrieved 2026-10-17T12:31:16Z; SHA-256 9289bc047b59e93f75703b72f5e92d9277e115238427519c343df550220e351b)
- https://export.arxiv.org/api/query?id_list=2602.19870%2C2602.22769%2C2602.23166%2C2602.23320%2C2603.00198%2C2603.00842%2C2603.00991%2C2603.01050%2C2603.01145%2C2603.02236%2C2603.02795%2C2603.03278%2C2603.04549%2C2603.06713%2C2603.07106%2C2603.07670%2C2603.07853%2C2603.08706%2C2603.09716%2C2603.10002%2C2603.10664%2C2603.11139%2C2603.11320%2C2603.12483%2C2603.13026%2C2603.13247%2C2603.15714%2C2603.16065%2C2603.16289%2C2603.16839%2C2603.17216%2C2603.17378%2C2603.19313%2C2603.19685%2C2603.21489%2C2603.22527%2C2603.23638%2C2603.24157%2C2603.24639%2C2603.27646%2C2603.28069%2C2603.28407%2C2603.29418%2C2603.29847%2C2604.00528%2C2604.00886%2C2604.01194%2C2604.01438%2C2604.01560%2C2604.02985&max_results=50 (retrieved 2026-10-17T12:31:16Z; SHA-256 9289bc047b59e93f75703b72f5e92d9277e115238427519c343df550220e351b)