Jingwei Ge

Jingwei Ge

Auto-generated profile, based on publicly available information

Jingwei Ge

Department of Automation, BNRist, Tsinghua University, Beijing, China

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
13 May 2024 in IEEE Transactions on Systems, Man, and Cybernetics: Systems
Verified
Journal Article
Closed access
CCCheng Chang
Cheng Chang
SWSiqi Wang
Siqi Wang
JZJiawei Zhang
Jiawei Zhang
JGJingwei Ge
Jingwei Ge
LLLi Li
Li Li
Scenario engineering plays a vital role in various Industry 5.0 applications. In the field of autonomous driving systems, driving scenario data are important for the training and testing of critical modules. However, the corner scenario cases are usually rare and necessary to be extended. Existing methods cannot handle the interpretation and reasoning of the generation process well, which reduces the reliability and usability of the generated scenarios. With the rapid development of Foundation Models, especially the large language model (LLM), we can conduct scenario generation via more powerful tools. In this article, we propose LLMScenario, a novel LLM-driven scenario generation framework, which is composed of scenario prompt engineering, LLM scenario generation, and evaluation feedback tuning. The minimum scenario description specific to LLM is given by scenario analysis and ablation studies. We also appropriately design the score functions in terms of reality and rarity to evaluate the generated scenarios. The model performance is further enhanced through chain-of-thoughts and experiences. Different LLMs are also compared with our framework. Experimental results on naturalistic datasets demonstrate the effectiveness of LLMScenario, which can provide solid support for scenario engineering in Industry 5.0.

13 May 2024 in IEEE Transactions on Intelligent Vehicles
Verified
Journal Article
Closed access
JGJingwei Ge
Jingwei Ge
CCCheng Chang
Cheng Chang
JZJiawei Zhang
Jiawei Zhang
LLLingxi Li
Lingxi Li
XNXiaoxiang Na
Xiaoxiang Na
FWFei-Yue Wang
Fei-Yue Wang
The deployment of large language models (LLM) brings challenges to intelligent systems because its capability of integrating large-scale training data facilitates contextual reasoning. This paper envisions a revolution of the LLM based (Artificial) Intelligent Operating Systems (IOS, or AIOS) to support the core of automated vehicles. We explain the structure of this LLM-OS and discuss the resulting benefits and implementation difficulties.

16 January 2024 in IEEE Transactions on Intelligent Transportation Systems
Verified
Journal Article
Closed access
JGJingwei Ge
Jingwei Ge
JZJiawei Zhang
Jiawei Zhang
CCCheng Chang
Cheng Chang
YZYi Zhang
Yi Zhang
DYDanya Yao
Danya Yao
LLLi Li
Li Li
Sampling, generation, and evaluation of scenarios are essential steps for intelligent testing of autonomous vehicles. Since uncertainty in driving behavior always leads to different occurrence frequencies of scenarios, we have to sample these scenarios in naturalistic datasets. Furthermore, a specified scenario needs to be further enriched and the driving behavior within it needs to be fully described to carry out generation in simulation systems. However, existing approaches generate scenarios randomly and uncontrollably, which makes them unable to precisely generate the specified scenarios. The driving behavior they describe is also memoryless and inflexible. To address the two issues, we propose a task-driven controllable scenario generation framework that can generate scenarios with the consideration of the driving behavior of Surrounding Vehicles (SVs) in a controllable manner. We first manually assign the driving behavior based on different testing tasks for all the considered vehicles. Then we expand the driving behavior temporally as the continuation and transition of several motion activities and generate the corresponding vehicle trajectories spatially. We adopt And-Or Graph (AOG) to model the transition between these motion activities. In contrast to the common memoryless Markov process, our framework generates driving behavior with continuity and driving memory. Finally, we evaluate our framework by generating lane-changing scenarios.

13 May 2024 in IEEE Transactions on Intelligent Vehicles
Verified
Journal Article
Closed access
CCCheng Chang
Cheng Chang
JZJiawei Zhang
Jiawei Zhang
JGJingwei Ge
Jingwei Ge
ZZZuo Zhang
Zuo Zhang
JWJunqing Wei
Junqing Wei
FWFei-Yue Wang
Fei-Yue Wang
Intelligent vehicles and autonomous driving systems rely on scenario engineering for intelligence and index (I&I), calibration and certification (C&C), and verification and validation (V&V). To extract and index scenarios, various vehicle interactions are worthy of much attention, and deserve refined descriptions and labels. However, existing methods cannot cope well with the problem of scenario classification and labeling with vehicle interactions as the core. In this paper, we propose VistaScenario framework to conduct interaction scenario engineering for vehicles with intelligent systems for transport automation. Based on the summarized basic types of vehicle interactions, we slice scenario data stream into a series of segments via spatiotemporal scenario evolution tree. We also propose the scenario metric Graph-DTW based on Graph Computation Tree and Dynamic Time Warping to conduct refined scenario comparison and labeling. The extreme interaction scenarios and corner cases can be efficiently filtered and extracted. Moreover, with naturalistic scenario datasets, testing examples on trajectory prediction model demonstrate the effectiveness and advantages of our framework. VistaScenario can provide solid support for the usage and indexing of scenario data, further promote the development of intelligent vehicles and transport automation.

16 August 2023 in IEEE Transactions on Intelligent Vehicles
Verified
Journal Article
Closed access
JZJiawei Zhang
Jiawei Zhang
JGJingwei Ge
Jingwei Ge
SLShu Li
Shu Li
SLShen Li
Shen Li
LLLi Li
Li Li
Cooperative driving of connected and automated vehicles (CAVs) has attracted extensive attention and researchers have proposed various approaches. However, existing approaches are limited to small-scale isolated scenarios and gaps remain in network-wide cooperative driving, especially in routing. In this paper, we decompose the network-level cooperative driving problem into two dominant sub-problems and accordingly propose a bi-level network-wide cooperative driving approach. The dynamic routing problem is considered in the upper level and we propose a multi-agent deep reinforcement learning (DRL) based routing model. The model can promote the equilibrium of network-wide traffic through distributed self-organized routing collaboration among vehicles, thereby improving efficiency for both individual vehicles and global traffic systems. In the lower level, we focus on the right-of-way assignment problem at signal-free intersections and propose an adaptive cooperative driving algorithm. The algorithm can adaptively evaluate priorities of different lanes, and then uses the lane priorities to guide the Monte Carlo tree search (MCTS) for better right-of-way assignments. Essentially, the upper level determines which conflict areas the vehicles will pass through, and the lower level addresses how the vehicles use the limited road resources more efficiently in each conflict area. The experimental results show that the upper and lower levels complement each other and work together to significantly improve the network-wide traffic efficiency and reduce the travel time of individual vehicles. Moreover, the results demonstrate that microscopic and mesoscopic cooperative driving behaviors of vehicles can significantly benefit the macroscopic traffic system.

26 January 2024 in IEEE Transactions on Intelligent Vehicles
Verified
Journal Article
Closed access
JGJingwei Ge
Jingwei Ge
JZJiawei Zhang
Jiawei Zhang
CCCheng Chang
Cheng Chang
YZYi Zhang
Yi Zhang
DYDanya Yao
Danya Yao
LLLi Li
Li Li
The substantial resource usage required to create ample scenarios for testing Autonomous Vehicles (AV) presents a bottleneck in their implementation. At present, research relies on sampling the driving behaviour of Surrounding Vehicles (SV) based on naturalistic datasets in simulation. However, these methods still generate huge amounts of scenarios, making it impossible to synthetically evaluate AV intelligence in a very small number of tests (especially in real-world situations). Simultaneously, the unknown distribution of critical scenarios leads to the problem that more critical scenarios cannot be accurately sampled. In this paper, a novel optimization problem is described and a dynamic scenario sampling method is proposed to cover more critical scenarios with finite samples. First, the sampling space is constructed by extracting the behavioural model parameters of the SVs. Second, multiple rounds of sampling are carried out successively to learn the distribution of critical scenarios, which in turn gradually improves the coverage of the critical scenarios. To do this, in each round, we divide the sampling space into several subspaces using two-step sampling, sample the scenarios using Random Quasi Monte Carlo (RQMC), evaluate the criticality of the subspace, and then use the evaluation results to guide the selection of the sampling space for the next round. The purpose of RQMC is to uniformly sample in the critical subspace rather than Standard Monte Carlo (SMC). Experimental results show that our method can better narrow the gap with the distribution of critical scenarios and discover more critical scenarios when compared to the baseline method.