Tingting Du
🎓 About Me
I am a B.S. student in Computer Science and Mathematics at the University of Wisconsin-Madison. Before that, I was a visiting student in Computer Science at UC Berkeley, and received training in Linguistics at Ningbo University.
I have been fortunate to collaborate with researchers at University of Maryland, College Park (working with Prof. Ang Li on Vision-Language-Action models), University of Notre Dame (working with Prof. Meng Jiang on student modeling and question generation), and Berkeley Artificial Intelligence Research (working with Prof. Alane Suhr on situated language understanding).
I am broadly interested in building intelligent robotic systems that can understand and interact with the physical world. My research focuses on:
- Vision-Language-Action (VLA) models for robotic manipulation and control
- Large Language Model reasoning and interpretability
- Multimodal learning bridging vision, language, and action
🔬 Research Interests
My research focuses on developing collaborative, efficient, and trustworthy machine learning systems for robotics, spanning:
Vision-Language-Action Models
Developing architectures that bridge 2D visual perception, 3D spatial understanding, and robotic action generation. My work on multi-layer shared projector frameworks aims to improve spatial reasoning in VLA systems.
Selected work:
- ROCKET (arXiv 2026): Residual-Oriented Multi-Layer Alignment for Spatially-Aware Vision-Language-Action Models
- VLA Survey (TMLR 2026): Comprehensive analysis of datasets, benchmarks, and data engines for Vision-Language-Action in robotics
LLM Reasoning & Interpretability
Understanding how large language models correct logical errors during training, using token-level analysis to pinpoint where and how reasoning improvements emerge.
Selected work:
- Developed interpretability frameworks using token-level KL-divergence for RLVR training analysis
- Published work on question generation and student modeling at ACL 2025
Grounded Language Understanding
Analyzing how humans develop ad-hoc conventions in collaborative environments, informing more natural human-AI interaction.
Selected work:
- Characterized language use in collaborative situated games (arXiv 2025)
- Agent KB: Hierarchical memory framework for cross-domain problem solving (ICML 2025 Workshop)
📰 Recent News
- Apr 2026: Our VLA survey paper accepted to TMLR!
- Feb 2026: ROCKET paper on arxiv, manuscript in submission to ICML
- Jan 2025: QG-SMS paper accepted to ACL 2025!
- Dec 2024: Characterized language use paper on arxiv
- Jun 2024: Completed research internship at Berkeley AI Research
📫 Contact
Feel free to reach out for research discussions or collaborations!
Email: tdu35 [at] wisc [dot] edu
You can also find me on GitHub, LinkedIn, and Google Scholar.