黄苏恺 | Sukai Huang

Postdoctoral Research Fellow, Monash University

About Me

I am a Postdoctoral Research Fellow at Monash University, working on the DARPA HARNESS and ONR SEA-AI projects under A/Prof. Hamid Rezatofighi (Lead PI, DARPA HARNESS; Area Chair, CVPR/NeurIPS/IROS) in the Vision & Learning for Autonomous AI (VL4AI) Group, with a focus on neuro-symbolic AI, large language models for embodied agents, and robot planning & reasoning.

I completed my PhD at the University of Melbourne (2021–2025), supervised by A/Prof. Nir Lipovetzky (ICAPS Executive Council Mentoring Chair, 2025–2031) and Prof. Trevor Cohn (Research Scientist, Google Research Australia), where my thesis investigated Integrating Natural Language for Sequential Decision Problems. Prior to that, I earned a Bachelor of Advanced Computing (Honours) from the Australian National University, supervised by Prof. Jochen Renz (ARC Future Fellow; Former Head of ANU AI Group).

Research Vision: I believe the next wave of AI will be agentic and verifiable systems that plan, act, and explain their decisions by composing neural perception with symbolic representations and logic. My work bridges the gap between probabilistic language models and formal reasoning, aiming to build embodied agents that are safe-by-default, auditable, and robust.

🔥 News

Jan 2026 — Paper accepted at AAMAS 2026 (First Author): Neurosymbolic Active Goal Recognition in Partially Observable Environments.
Jan 2026 — Paper accepted at ACL 2026 (Corresponding Author): ALIGN: Word Association Learning for Cultural Alignment in LLMs.
Jan 2026 — Paper accepted at ICLR 2026: MATA: A Trainable Hierarchical Automaton System for Multi-Agent Visual Reasoning.
Jun 2025 — Started Postdoc at Monash University; joined DARPA HARNESS and ONR SEA-AI projects as a core member.
Jun 2025 — Invited to give a presentation at the Kingston AI Symposium 2026 in Adelaide, Australia.
Jun 2025 — ICAPS 2025 Oral Presentation: Chasing Progress, Not Perfection: Revisiting Strategies for End-to-End LLM Plan Generation.
Mar 2025 — Two oral presentations at the LM4Plan Workshop @ AAAI 2025.

Selected Publications

Sorted by contribution significance. First-author / corresponding-author papers are highlighted.

【First Author】 Sukai Huang, Chenyuan Zhang, Hamid Rezatofighi, Mor Vered, and Buser Say. “Neurosymbolic Active Goal Recognition in Partially Observable Environments.” International Conference on Autonomous Agents and Multi-Agent Systems (AAMAS), 2026.
【First Author】 Sukai Huang, Nir Lipovetzky, and Trevor Cohn. “Chasing Progress, Not Perfection: Revisiting Strategies for End-to-End LLM Plan Generation.” International Conference on Automated Planning and Scheduling (ICAPS), 2025. 🏷️ Oral Presentation
【First Author】 Sukai Huang, Nir Lipovetzky, and Trevor Cohn. “Planning in the Dark: LLM-Symbolic Planning Pipeline without Experts.” AAAI Conference on Artificial Intelligence (AAAI), 2025.
【Corresponding Author】 Chunhua Liu, Kabir Manandhar Shrestha, and Sukai Huang. “ALIGN: Word Association Learning for Cultural Alignment in Large Language Models.” Annual Meeting of the Association for Computational Linguistics (ACL), 2026.
Zhixi Cai, Fucai Ke, Kevin Leo, Sukai Huang, Maria Garcia de la Banda, Peter J. Stuckey, and Hamid Rezatofighi. “MATA: A Trainable Hierarchical Automaton System for Multi-Agent Visual Reasoning.” International Conference on Learning Representations (ICLR), 2026.
Simon De Deyne, Sukai Huang, Lea Frermann, and Chunhua Liu. “Fine-tuned Large Language Models Predict Human Word Associations and Improve Performance across Lexical and Semantic Processing Tasks.” Annual Meeting of the Cognitive Science Society (CogSci), 2026.
【First Author】 Sukai Huang, Shu-Wei Liu, Nir Lipovetzky, and Trevor Cohn. “The Dark Side of Rich Rewards: Understanding and Mitigating Noise in VLM Rewards.” ICAPS 2025 Workshop on Bridging the Gap Between AI Planning and Reinforcement Learning (PRL). [arXiv]

Pre-prints & Code: [Google Scholar] | [GitHub]

Research Projects

DARPA HARNESS Project

Hierarchical Abstractions and Reasoning for Neuro-Symbolic Systems

Role: Research Fellow
Period: 2025 – Present
Focus: Developing neuro-symbolic planning and granularity-aware instruction-following modules for autonomous robotic systems, integrating perception, reasoning, and planning through neuro-symbolic frameworks.

ONR SEA-AI Project

Neuro-Symbolic Enhanced Autonomy for Maritime Scene Understanding

Role: Research Fellow
Period: 2025 – Present
Focus: Leading mission analysis from natural language, world-model construction, active perception, and sequential decision-making components for maritime autonomous systems.

Academic Services

Reviewer: ICAPS, ACL Rolling Review (ARR), and the Artificial Intelligence journal.
Workshop Organizer / PC Member: Active contributor to the neuro-symbolic AI and LLM-planning communities.

Awards & Honors

AAAI 2025 Scholarship, Association for the Advancement of Artificial Intelligence.
ICAPS 2025 Scholarship, International Conference on Automated Planning and Scheduling.
Kingston AI Symposium 2026 — Invited Presenter, Adelaide, Australia.
Winner, Aesthetic Track of the Angry Birds Level Generation Competition, IEEE COG 2020.
ANU Summer Scholarship Program 2019 — Researcher on emotion recognition models.
ANU Chancellor’s Letters of Commendation (2018, 2020).
ANU Dean’s Award (2019).

Teaching & Industry Experience

Tutor (Feb 2020 – Jul 2021), Australian National University
Planned and delivered interactive tutorials on Python Programming and Data Management for undergraduate students.
Machine Learning Engineer (May 2022 – Nov 2023), FINVISE London (Remote Contract)
Built an Automated Valuation Model (AVM) achieving >80% accuracy; designed a multimodal data pipeline handling 20M+ property records; proposed Masked Autoencoder (MAE) solutions for missing-data challenges.

Education

PhD in Engineering & IT — University of Melbourne, Australia, 2021–2025
Thesis: Integrating Natural Language for Sequential Decision Problems
Scholarship: Melbourne Research Scholarship
Supervisors: A/Prof. Nir Lipovetzky (ICAPS Executive Council Mentoring Chair, 2025–2031) & Prof. Trevor Cohn (Research Scientist, Google Research Australia)
Bachelor of Advanced Computing (Honours) — Australian National University, Australia, 2017–2021
Honours: First Class Honours | GPA: 6.84 / 7.00
Specialization: Machine Learning Systems
Supervisor: Prof. Jochen Renz (ARC Future Fellow; Former Head of ANU AI Group)

Contact

I am open to academic collaborations and faculty opportunities. Please feel free to reach out!

Email: huangsukai1997@gmail.com
LinkedIn: Sukai Huang
GitHub: Sino-Huang
Google Scholar: Profile
Personal Homepage: https://sino-huang.github.io/

Full CV

Page: /

My Research

The Big Picture of My Research Integrating natural language into current AI system is a promising direction to democratize AI technology. Moreover, the vast knowledge embedded in natural language presents an opportunity to enhance AI-driven decision-making. Imagine you want to instruct an AI system in Minecraft to build a house. Instead of programming a detailed set of construction rules or crafting reward functions that require expert insight, you simply tell the AI: “Build a two-story house with a garden, using bricks for the walls and wood for the roof.” A natural language (NL)-integrated sequential decision making (SDM) system shall leverage its understanding of natural language to break down this instruction into a series of actionable steps, and progressing to the desired goal you want! Problem (TL;DR) The current design of these natural language-integrated AI systems has significant room for improvement. For example, the algorithms often lack robustness and efficiency, which undermines the reliability of sequential decision making. Know more about the problem setting (two paradigms for sequential decision-making) There are two primary paradigms for sequential decision-making: imagine you are playing Minecraft — a complex, open-ended problem-solving environment where players can build, explore, and survive. There are two ways you might approach this task. Click here to know more about the context My research during PhD study: Improving NL-integrated SDM systems under the two paradigms 1. Model-free reinforcement learning (RL) -- (a) VLM + Language-based Reward + RL agent) Problem a: Noisy rewards from language models misguide AI agents. Solution: BiMI Reward Function (paper) Reduces false positives (e.g., rewarding irrelevant actions). Combines mutual information and thresholding for robustness. Result: Faster learning in navigation tasks (e.g., robots avoiding obstacles). 2. Model-based automated planning -- (b) LLM-Symbolic Planning Pipeline and (c) LLMs for plan generation) Problem b: LLMs hallucinate plans or require expert validation. Solution: Fully Automated LLM-Symbolic Pipeline (paper) Generates and validates action schemas without human intervention. Resolves ambiguity by exploring multiple interpretations of language. Result: Outperforms expert-dependent methods in scalability and bias reduction. Problem c: There has been ongoing controversy about the genuine planning abilities of LLMs, with critics questioning whether their outputs reflect true reasoning or superficial statistical patterns. Contribution: Reassessment of LLMs for end-to-end plan generation (paper) Conducts a rigorous re-evaluation of various strategies claiming to enhance LLM reasoning in end-to-end planning, using diverse metrics for a comprehensive assessment. Found that RL promotes better generalization than supervised fine-tuning (SFT) for training LLMs to plan ...

黄苏恺 | Sukai Huang

About Me#

🔥 News#

Selected Publications#

Research Projects#

DARPA HARNESS Project#

ONR SEA-AI Project#

Academic Services#

Awards & Honors#

Teaching & Industry Experience#

Education#

Contact#

Full CV#