Skip to content
Tongji Alumna Joins Rutgers to Pioneer Physics-Enhanced AI for Autonomous Driving

Tongji Alumna Joins Rutgers to Pioneer Physics-Enhanced AI for Autonomous Driving

Tongji alumna Long Keke will join Rutgers University in 2027 as an assistant professor, advancing physics-enhanced AI for autonomous driving through methods like physics-infused trajectory prediction, vision-language model control, and hallucination mitigation for safer real-world AI deployment.

SmartApe AI
SmartApe AI
SmartApe AI

Background and Appointment

Long Keke, a Tongji University alumna (master's degree) and former postdoctoral researcher at the University of Wisconsin–Madison, will join Rutgers University–New Brunswick's School of Engineering as an assistant professor in January 2027. She earned her Ph.D. in Civil and Environmental Engineering from UW–Madison in 2024. Her research centers on applying AI to real-world transportation environments, specifically physics-enhanced AI, autonomous driving, and intelligent transportation systems.

Physics-Enhanced Residual Learning for Trajectory Prediction

In early work, Long sought to embed physical laws into machine learning so that AI systems obey real-world motion constraints rather than relying solely on data. She proposed Physics-Enhanced Residual Learning (PERL) for vehicle trajectory prediction. The method first uses a traditional physics-based model to generate a baseline prediction, then employs a neural network to learn and correct the residual errors. This hybrid approach was published in Communications in Transportation Research in 2025.

PERL method illustration
PERL method illustration

Vision-Language Model Assisted Autonomous Driving Control

Extending from prediction to decision-making, Long and collaborators introduced VLM-MPC in 2026, a framework that couples a vision-language model (VLM) with a conventional model predictive control (MPC) system. The VLM interprets complex road scenes and provides high-level driving suggestions, while the MPC executes low-level vehicle control actions. This separation reduces safety risks associated with end-to-end direct control by large models. The system was evaluated in the CARLA simulator across challenging scenarios including night driving, rain, and intersections.

VLM-MPC architecture
VLM-MPC architecture

Mitigating Hallucinations in Transportation Scene Understanding

Recognizing that VLMs can generate plausible but incorrect descriptions of traffic scenes, Long co-authored a study on hallucination mitigation accepted by Transportation Research Part C in October 2026. The approach compares multiple model outputs for the same scene, identifies inconsistencies, and filters for more reliable descriptions. On the nuScenes dataset , the method achieved an 87% accuracy in selecting hallucination-free descriptions.

Hallucination mitigation results
Hallucination mitigation results

Cooperative Perception Dataset for Vehicle-to-Vehicle Communication

Long also contributed to the CATS-V2V dataset , a vehicle-to-vehicle cooperative perception benchmark accepted at CVPR 2026 . The dataset supports research on information sharing and collaborative perception among autonomous vehicles.

CATS-V2V dataset overview
CATS-V2V dataset overview

Professional Roles and Honors

Co-Principal Investigator on U.S. federally funded transportation research projects.

Instructor for the Center for Connected and Automated Transportation (CCAT).

Associate Editor of Expert Systems .

Awards: WISL Science Communication Award, NGTS-3 Best Presentation Award, CUTR Student Poster Competition First Prize, National Scholarship (China), Shanghai Outstanding Graduate.

Future Research Directions at Rutgers

At Rutgers, Long plans to further explore vision-language models, vision-language-action (VLA) models, and world models. Her goal is to move AI from understanding images and traffic scenes toward robust perception, decision-making, and action in the physical world, expanding beyond intelligent transportation into broader physical-world intelligence.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

vision-language modelshallucination mitigationautonomous drivingtrajectory predictionCARLA simulatorcooperative perceptionnuScenes datasetphysics-enhanced AI
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.