Augusto B. Corrêa
Junior Research Fellow · Christ Church · University of Oxford
I am a Junior Research Fellow at Christ Church, University of Oxford. My current interest is how LLMs can automate research, with a focus on automated planning. I study this through self-improvement and interaction with formal systems such as planners and proof assistants.
My recent work shows that LLMs can design state-of-the-art heuristics for planning domains (NeurIPS 2025, NeurIPS 2026), and that coding agents can improve a planner on their own. My research was also featured in the Stanford AI Index Report 2026.
Before Oxford, I obtained my Ph.D. and M.Sc. at the University of Basel, Switzerland, supervised by Prof. Malte Helmert. My Ph.D. thesis received the ICAPS Outstanding Dissertation Award (2025). I created the Powerlifted planning system, and my planners Scorpion Maidu and Levitron won the 2023 International Planning Competition.
For more details, see my academic CV or its shorter version.
News
| Sep 26, 2026 | Two papers accepted at NeurIPS 2026: Hierarchical Task Network Planning with LLM-Generated Heuristics and Property-Guided LLM Program Synthesis for Planning. |
|---|---|
| Aug 01, 2026 | Our paper Frontier Large Language Models Are Becoming Competitive with State-of-the-Art Planners was accepted at Communications of the ACM. |
| Jun 01, 2026 | Our paper Potential Heuristics as Real-Valued Multilinear Polynomials was selected as runner-up for the Best Paper Award at ICAPS 2026. |
| May 27, 2026 | Two new preprints are available. Check the publications page! |
| Feb 20, 2026 | Paper accepted at ICAPS 2026: Potential Heuristics as Real-Valued Multilinear Polynomials. |
| Feb 19, 2026 | Awarded the Google Gemini Academic Program Award for research on LLMs for planning. |
| Dec 30, 2025 | New preprint: Iterative Deployment Improves Planning Skills in LLMs. |
| Oct 14, 2025 | Awarded the ICAPS Outstanding Dissertation Award for my Ph.D. thesis. |
| Sep 18, 2025 | Paper accepted at NeurIPS 2025: Classical Planning with LLM-Generated Heuristics: Challenging the State of the Art with Python Code. |