Han Luo
I am a third-year undergraduate student in Computer Science at the University of Leeds, UK, and Southwest Jiaotong University, China. My research focuses on reliable agents and recursive self-improvement (RSI).
I am fortunate to work with Prof. Lucy Lu Wang in the LARCH Lab at the University of Washington. Previously, I worked with Prof. Guy Laban in LabaLab at
Ben-Gurion University of the Negev and at the
University of Cambridge.
NeurIPS 2026 · Sydney 🇦🇺
I'll be there! Open to research internships, collaborations, and coffee chats on reliable agents and RSI. Let's connect
My research explores two closely connected directions:
Reliable Agents
- Knowing when to act and when to abstain. Agentic Abstention (NeurIPS 2026) · HERA
- Maintaining behavioral consistency and evaluating safety in agent interactions. SPASM (ACL 2026) · DialogGuard (ACL 2026)
- Transferring learned experience to open-world embodied navigation. SAGE (ICML 2026)
Recursive Self-Improvement (RSI)
- Evolving agent context through lessons learned from interaction trajectories. CONVOLVE (NeurIPS 2026)
- Co-evolving agent harnesses and environments through execution feedback. HERA
Selected publications
View all >HERA
Co-evolving agent harnesses and environments for reliable agentic abstention.
Agentic Abstention
A benchmark and context-engineering study of when LLM agents should stop acting and abstain.
Plan in Sandbox
A physics-grounded sandbox framework for learning abstracted embodied navigation experience.
SPASM
A stability-first framework for persona-driven LLM-LLM dialogue simulation with egocentric context projection.
DialogGuard
A practitioner-facing interface for multi-agent psychosocial safety evaluation of sensitive LLM responses.
News
-
Our HERA preprint on harness–environment co-evolution is now available on arXiv.
-
Agentic Abstention has been accepted to NeurIPS 2026. See ya in Sydney 🇦🇺!
-
I spent two months at LARCH Lab, University of Washington, in Seattle 🇺🇸, working with Prof. Lucy Lu Wang.
-
Plan in Sandbox, Navigate in Open Worlds has been accepted to ICML 2026. See ya in Seoul 🇰🇷!
-
SPASM has been accepted to Findings of ACL 2026. See ya in San Diego 🇺🇸!
-
DialogGuard has been accepted to ACL 2026 System Demonstrations. See ya in San Diego 🇺🇸!
-
I spent two months at LabaLab, Ben-Gurion University of the Negev, in Israel 🇮🇱, working with Prof. Guy Laban.
-
My Cambridge OSRP project received a Best Research Project nomination.