Hung-Ting Su
Research Fellow, National University of Singapore
Ph.D. in Computer Science, National Taiwan University. Previously an industry researcher in robot learning at Delta Electronics and a visiting researcher at Columbia University.
At NUS, I work with Prof. Hwee Tou Ng to study how multimodal AI can detect harmful video content and remain reliable when adversarial inputs attempt to manipulate its behavior.
My broader research connects trustworthy LLM/MLLM reasoning, long-horizon multimodal evaluation, and real-world robot learning.
Previously, I was a Senior Associate Researcher at Delta Robotics Innovation Center, Delta Electronics, working on tactile/visual robot learning, VLA-style manipulation, and rollout evaluation.
Earlier, I completed my Ph.D. and later postdoctoral research at National Taiwan University under the supervision of Prof. Winston H. Hsu. During that time, I also collaborated closely with Prof. Min Sun, Prof. Hung-yi Lee, and Prof. Pu-Jen Cheng on trustworthy multimodal reasoning and embodied AI. I was also a visiting researcher at Columbia University, hosted by Prof. Shih-Fu Chang.
Selected Publications
|
VLN-NF: Feasibility-Aware Vision-and-Language Navigation with False-Premise Instructions
Reliable embodied reasoning when instructions contain invalid assumptions.
|
|
Unveiling Narrative Reasoning Limits of Large Language Models with Trope in Movie Synopses
Evaluating narrative reasoning limits and failure modes in LLMs.
|
|
Context-Aware Replanning with Pre-Explored Semantic Map for Object Navigation
Semantic replanning for more robust embodied navigation.
|
|
ADAPT: Benchmarking Commonsense Planning under Unspecified Affordance Constraints
Benchmarking commonsense planning under underspecified affordance constraints.
|
|
Affordance-Guided Coarse-to-Fine Exploration for Base Placement in Open-Vocabulary Mobile Manipulation
Affordance-aware exploration for mobile manipulation and base placement.
|
|
MovieCORE: COgnitive REasoning in Movies
Long-form multimodal reasoning over movie narratives.
|
|
HERMES: temporal-coHERent long-forM understanding with Episodes and Semantics
Temporal coherence and semantic structure for long-horizon movie understanding.
|
|
AED: Adaptable Error Detection for Few-shot Imitation Policy
Monitoring and error detection for imitation policies in embodied settings.
|
Research Interests
- Multimodal AI safety and adversarial robustness
- Real-world robot learning and embodied AI
- Trustworthy LLM and MLLM reasoning
- Long-horizon multimodal understanding and evaluation
- Tactile/VLA-style manipulation, world models, and failure analysis
About
My academic research spans trustworthy LLM/MLLM reasoning, long-form video understanding, and multimodal evaluation, developed through work at NTU and Columbia and now extended at NUS to harmful video content detection and adversarial robustness.
Across robotics, language, and vision, I study how multimodal systems gather evidence, revise beliefs, and remain reliable when observations are incomplete, ambiguous, or physically constrained.
Academic Service
Area Chair: NAACL/COLING '27 (ARR 10/26); ACL '26 (ARR 01/26); EACL '26 (ARR 10/25).
Reviewer: ICML, NeurIPS, CVPR, ICCV, NAACL, ACL, EMNLP, CoRL, ICRA, IROS, AAAI, ACM Multimedia, TCSVT.
