Han Zheng — MIT

News

2026-10-15
Invited to present CORAL at the 2026 INFORMS Annual Meeting. See you in SF. Feel free to reach out for coffee chat.
2026-10-08
Presented CORAL at COLM 2026 with Ao and Zijian.
2026-09-28
Reef reached 7,000 stars on GitHub.
2026-09-24
Three papers accepted to NeurIPS 2026: D-PACE (Spotlight), Economy of Minds, and FrontierOR.
2026-09-02
We open-sourced Reef, an infrastructure for continually self-improving agents. Code / Post on X

Show moreShow less

2026-07-08
CORAL accepted to COLM 2026.
2026-06-01
Started a research internship at Amazon AGI, working on agentic post-training.
2026-05-28
Presented CORAL at the Jump Trading AI Symposium.
2026-04-02
We open-sourced CORAL, the first autonomous multi-agent framework for open-ended discovery. Post on X
2026-03-24
Our paper on RL-guided multi-robot planning was accepted to JAIR. Read more in MIT News and the Symbotic blog.
2026-02-15
AlphaOPT was accepted to KDD 2026.
2025-06-01
Started a research internship at Symbotic, working on learning-guided multi-robot path planning.

Research

I'm interested in scalable planning and decision-making for multi-agent systems, combining reinforcement learning with classical search and combinatorial optimization. More recently, I've been exploring self-evolving agents that improve through interaction, accumulated experience, and open-ended discovery.

Internalizing Curriculum Judgment for LLM Reinforcement Fine-Tuning

H. Zheng, Y. Ma, K. Gunasekaran, B. Balaji, Z. Du, S. Vitaladevuni, C. Wu

arXiv, 2026

Introduced METIS, a self-evolving framework that internalizes curriculum judgment as a native LLM capability for RFT, letting the policy learn what to learn next and accelerating convergence by up to 67%.

D-PACE: Dynamic Position-Aware Cross-Entropy for Parallel Speculative Drafting

T. Wu, Y. Yao, Z. Qi, H. Zheng, Z. Wang, H. Ma, L. Liao, H. Lakkaraju, J. Li, Y. Du

NeurIPS, 2026 · Spotlight

Introduces D-PACE, a position-aware training objective that improves parallel speculative decoding speed and draft acceptance without modifying the model architecture or inference process.

Economy of Minds: Emerging Multi-Agent Intelligence with Economic Interactions

Z. Qi, H. Su, A. Qu, C. Wang, Y. Yao, H. Zheng, K. Chattopadhyay, G. Xu, Z. Wang, W. Ye, V. J. Reddi, J. Li, P. P. Liang, H. Lakkaraju, S. Kakade, Y. Du

NeurIPS, 2026

A decentralized multi-agent system inspired by Hayek's economics, where agents compete via auctions and evolve through economic selection. Simple market signals turn populations of weak partial agents into collective intelligence that outperforms stronger monolithic baselines across five agentic tasks.

CORAL: Towards Autonomous Multi-Agent Evolution for Open-Ended Discovery

A. Qu^*, H. Zheng^*, Z. Zhou^*, Y. Yan, Y. Tang, S. Y. Ong, F. Hong, K. Zhou, C. Jiang, M. Kong, J. Zhu, X. Jiang, S. Li, C. Wu, B. K. H. Low, J. Zhao, P. P. Liang

COLM, 2026

First framework for autonomous multi-agent evolution on open-ended problems. CORAL uses long-running LLM agents that explore, reflect, and collaborate through shared persistent memory and asynchronous execution, achieving 3–10× higher improvement rates over fixed evolutionary baselines across mathematical, algorithmic, and systems optimization tasks.

Learning-Guided Prioritized Planning for Lifelong Multi-Agent Path Finding in Warehouse Automation

H. Zheng, Y. Ma, B. Araki, J. Chen, C. Wu

JAIR, 2026

Developed RL-RH-PP, combining deep reinforcement learning with search-based prioritized planning for lifelong MAPF. Demonstrated superior throughput and strong generalization across agent densities and planning horizons.

AlphaOPT: Formulating Optimization Programs with Self-Improving LLM Experience Library

M. Kong, A. Qu, X. Guo, W. Ouyang, C. Jiang, H. Zheng, Y. Ma, D. Zhuang, Y. Tang, J. Li, H. Wang, C. Wu, J. Zhao

KDD, 2026 INFORMS Data Mining, 2025

Self-improving LLM framework that builds an experience library for formulating optimization programs across diverse problem domains.

Multi-Agent Path Finding for Mixed Autonomy Traffic Coordination

H. Zheng, Z. Yan, C. Wu

IROS, 2024

Pioneered BK-PBS, integrating offline-trained behavior prediction with multi-agent pathfinding to coordinate connected and human-driven vehicles under realistic kinematic constraints on highways.

Multi-Agent Path Finding for Cooperative Autonomous Driving

Z. Yan, H. Zheng, C. Wu

ICRA, 2024

Proposed OBS-KATS for signal-free intersection coordination. Proved soundness, completeness, and optimality of the crossing order search with significant throughput improvements.