Car-following and lane-changing models are cornerstones for understanding driver behavior, traffic dynamics, and supporting autonomous driving. As early as 2014, at the dawn of the deep learning era, my team proposed an innovative data-driven car-following model based on k Nearest Neighbors. This model has been cited more than 200 times, and it is widely recognized as one of the first data-driven car-following models by review articles published by scholars from institutions such as University of California, Berkeley (Zhang et al., 2025), University of Washington (Chen et al., 2023), Imperial College London (Rowam et al., 2025), and Tsinghua University (Li et al., 2026). In 2019, my team developed the first deep-learning-based lane-changing model using DBN and LSTM neural networks, which has since been cited over 300 times and is widely recognized in subsequent research (Siebke et al., 2023; Han et al., 2024; Wang et al., 2024; Huang et al., 2025) as a pioneering contribution to AI-driven modeling of driver behavior and traffic dynamics.
Prediction is a central topic in transportation research and practice. I contributed to the development of the first CNN-based traffic speed prediction model, and the single paper has been cited over 1,700 times. Other representative works include graph CNN-based bike-sharing demand prediction, GAN-based travel time prediction, etc. All these works were published in leading journals, such as TRC, CACAIE, and Information Fusion, and have received substantial citations, making a significant impact on the field of AI-driven transportation prediction.
An early innovator in large language model (LLM) applications in transportation. Since the emergence of LLMs in 2023, my team was among the first to introduce LLM-based intelligent driving assistance, highlight the positive role of prompt engineering in enhancing prediction accuracy, and develop AI agent models for everyday route choice, with a series of early papers published in leading journals. These works rank among the earliest applications of LLMs and have attracted wide attention in the research community. In 2024, I was twice invited to introduce those LLM-based progresses in ground mobility at SAE International, the world’s leading authority on automotive industry standards. At the beginning of 2026, we had a Correspondence, about AI writing tools could erase low-income countries‘ voices, which was published in Nature.
To address highway congestion and emissions, my team was the first to propose a feedback control–based congestion mitigation strategy for connected vehicles and the jam-absorption driving strategy, with results published in IEEE TITS and more than 300 citations and reported by Italian media. Researchers in Japan cited our work on jam-absorption strategies more than ten times in a single paper (Nishi, 2020), recognizing the two-step jam-absorption strategy as an effective means of alleviating traffic congestion. I led a survey paper co-authored with scholars from MIT, Georgia Tech, and TU Delft, which was published in IEEE TITS, reflecting my international influence in this topic.
For the urban scenarios, I introduced a novel future no-lane-change road transportation system for autonomous driving and subsequently extended the framework to include pedestrians. The series of works was published in IEEE TIV, IEEE TVT, and TRC, drawing wide attention from the research community (such as more than 100,000 views on LinkedIn). Researchers from institutions such as the University of Florida provided a detailed and positive assessment of the no-lane-change concept (Li et al., 2018), and subsequent studies by Antonio et al. (2022) and Zhao et al. (2025) considered it one of the first conflict-based intersection management.
I have also devoted significant effort to predicting the motion of surrounding agents for autonomous vehicles, with a particular focus on intersection conflict scenarios. To date, I have developed a series of methods, including approaches based on Transformer, LLMs with CoT reasoning, and World Models.
Time-space traffic diagrams are one of the most important traffic analysis and visualization tools. The series of works centered on the enhancement of the diagrams, including the mapping-to-cell approaches (highways, urban roads), the traffic emission estimation and the resolution enhancement (seminal, advanced). In particular, the wave-aware method (seminal, advanced) of constructing time-space diagrams has been substantively implemented in the well-known I-24 MOTION project in Nashville (Ji et al., 2024; Ji et al., 2026), as well as in the works of global researchers from the UK (Shang et al., 2023), Sweden (Tsanakas et al., 2022), and China (He et al., 2024). Notably, these implementations were carried out completely independently of our own research activities.
Representative works include the assessment of the dynamic vulnerability of transportation systems (seminal, multi-modal), a mesoscopic CO₂ emission estimation model with the consideration of traffic dynamics, multi-day rail-hailing behavior modeling, and in-lab experiment-based day-to-day route choice modeling (part I, part II), demonstrating my broad research interests, strong research skills, effective team leadership and collaboration, and wide-reaching impact.
I co-authored multiple highly cited review articles in transportation research, synthesizing trajectory data studies (citation > 300), automated vehicle-involved traffic (citation > 300), transportation system resilience (citation > 250), and also multi-modal resilience. These reviews have guided subsequent research, reflecting both scholarly recognition and field-wide impact.