LLM planning for VLA
Thinking about how LLM agents can plan, code, and act as the decision layer for vision-language-action systems.
PhD student, Technical University of Munich
Ruochen Li
I work on reliable clinical evaluation, LLM-based planning for VLA and agentic coding, and multimodal medical AI. My research asks how better evaluation can shape better models, especially when clinical judgment and real-world constraints matter.
I am a PhD student at RobUSt (Robotics and Ultrasound Team), Technical University of Munich, supervised by Prof. Nassir Navab and mentored by Dr. Yuan Bi. I completed my M.Sc. in Computer Science at TUM with distinction.
Current focus
What I am thinking about now
Reliable clinical evaluation
Building clinician-aligned evaluation signals that can expose model failures and guide better medical generation models.
Multimodal medical AI
Learning useful representations from report, WSI, CT/CTA, MRI, and raw medical signals that humans cannot easily inspect.
Recent news
Updates
-
PhD
Joined CAMP at TUM and started my PhD.
-
Collab
Developing agentic metrics for medical generated report evaluation with clinical alignment, in collaboration with Jean-Philippe Corbeil at Microsoft Healthcare.
-
Paper
ReEvalMed accepted to EMNLP 2025 Main.
-
Preprint
Beyond Scalar Scores released for LLM-based clinical significance evaluation.
-
Service
Reviewer for IEEE Transactions on Medical Imaging and volunteer at EMNLP 2025.
Research taste
A thread through the work
Reliable eval -> better models
I see evaluation as more than a leaderboard. When metrics reflect clinical judgment, they become useful training signals that pull models toward safer and more meaningful behavior.
Make model behavior legible
I care about opening the black box: understanding where models attend, why they fail, and how structural knowledge can make their decisions easier to audit.
Let AI read what humans cannot
From cardiac MRI k-space to CT radiomics, I like using AI to mine raw medical data for signals that are present but hard for humans to directly interpret.
Selected work
Publications and projects
ReEvalMed: Rethinking Medical Report Evaluation by Aligning Metrics with Real-World Clinical Judgment
A clinician-validated meta-evaluation benchmark for radiology report metrics, revealing clinical misinterpretations and score-inflating practices in widely used metrics.
Beyond Scalar Scores: Exploring LLM-based Metrics for Clinical Significance Evaluation in Radiology Reports
Post-trained Qwen-8B and MedGemma-4B with SFT/DPO + LoRA, outperforming 32B medical LLMs on clinical alignment while exposing evaluator bias and robustness failures.
Surgical Planning Evaluation for Video-LLMs
Studies how to evaluate long-horizon surgical planning in safety-critical settings, separating visual grounding failures from planning failures and examining how structural knowledge helps constrain model behavior.
Focus Directions Make Your Language Models Pay More Attention to Relevant Contexts
Training-free focus directions in key/query activations steer LLM attention toward task-relevant context, mitigating distraction across multiple LLM families.
Classification, Regression and Segmentation directly from k-Space in Cardiac MRI
Derived representations from UK Biobank k-space using masked autoencoders, enabling robust prediction from undersampled, human-unreadable frequency-domain data.
Multimodal CT Radiomics-Clinical Model for Hepatic Steatosis Risk Prediction with Metabolomic Interpretation
A multimodal model combining CT radiomics and clinical features for hepatic steatosis risk prediction, with metabolomics used to interpret imaging differences.
Learning Cascade Attention for Fine-Grained Image Classification
A weakly supervised parallel CNN architecture with spatial and cross-network attention mechanisms for fine-grained visual recognition.
Temporal Conditioning for Longitudinal Brain MRI Registration and Aging Analysis
A learning-based framework for sparse-scan longitudinal brain MRI registration and brain-state forecasting across aging trajectories.
Service
Academic service
- Reviewer, IEEE Transactions on Medical Imaging
- Reviewer, EMA4MICCAI 2025 Workshop
- Volunteer, EMNLP 2025
Contact
Get in touch
I am happy to have coffe chat.