Jiawei Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jiawei Wang has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Chong Guo
College of Automotive Engineering, Jilin University, Changchun, China
7 shared publications - Yao Zhang
College of Automotive Engineering, Jilin University, Changchun, China
7 shared publications - Liang Chu
Jilin University
3 shared publications - Yabin Mao
National Key Laboratory of Automotive Chassis Integration and Bionics, Jilin University, Changchun 130025, China
2 shared publications - Zhuo Cao
College of Automotive Engineering, Jilin University, Changchun, China
2 shared publications Chaoyi Chen
College of Automotive Engineering, Jilin University, Changchun, China
1 shared publication
Latest Publications
- Open Access
A Vision/Inertial Navigation/Global Navigation Satellite Integrated System for Relative and Absolute Localization in Land Vehicles
This paper presents an enhanced ground vehicle localization method designed to address the challenges associated with state estimation for autonomous vehicles operating in diverse environments. The focus is specifically on the precise localization of position and orientation in both local and global coordinate systems. The proposed approach integrates local estimates generated by existing visual–inertial odometry (VIO) methods into global position information obtained from the Global Navigation Satellite System (GNSS). This integration is achieved through optimizing fusion in a pose graph, ensuring precise local estimation and drift-free global position estimation. Considering the inherent complexities in autonomous driving scenarios, such as the potential failures of a visual–inertial navigation system (VINS) and restrictions on GNSS signals in urban canyons, leading to disruptions in localization outcomes, we introduce an adaptive fusion mechanism. This mechanism allows seamless switching between three modes: utilizing only VINS, using only GNSS, and normal fusion. The effectiveness of the proposed algorithm is demonstrated through rigorous testing in the Carla simulation environment and challenging UrbanNav scenarios. The evaluation includes both qualitative and quantitative analyses, revealing that the method exhibits robustness and accuracy.
Open AccessA Vision/Inertial Navigation/Global Navigation Satellite Integrated System for Relative and Absolute Localization in Land Vehicles
This paper presents an enhanced ground vehicle localization method designed to address the challenges associated with state estimation for autonomous vehicles operating in diverse environments. The focus is specifically on the precise localization of position and orientation in both local and global coordinate systems. The proposed approach integrates local estimates generated by existing visual–inertial odometry (VIO) methods into global position information obtained from the Global Navigation Satellite System (GNSS). This integration is achieved through optimizing fusion in a pose graph, ensuring precise local estimation and drift-free global position estimation. Considering the inherent complexities in autonomous driving scenarios, such as the potential failures of a visual–inertial navigation system (VINS) and restrictions on GNSS signals in urban canyons, leading to disruptions in localization outcomes, we introduce an adaptive fusion mechanism. This mechanism allows seamless switching between three modes: utilizing only VINS, using only GNSS, and normal fusion. The effectiveness of the proposed algorithm is demonstrated through rigorous testing in the Carla simulation environment and challenging UrbanNav scenarios. The evaluation includes both qualitative and quantitative analyses, revealing that the method exhibits robustness and accuracy.
- 8 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25 (9) , 12203-12215
A Sequence-to-Sequence Car-Following Model for Addressing Driver Reaction Delay and Cumulative Error in Multi-Step Prediction
Car-following behavior is one of the most common driving behaviors. To reduce the impact of driver reaction delay and accumulated errors in predicting long sequences on the accuracy of speed prediction, we propose a deep learning car-following model based on a sequence-to-sequence (seq2seq) architecture with an attention mechanism. Firstly, we analyze the characteristics of driver reaction delay during the car-following process and design an attention mechanism to learn the probability distribution of driver reaction delay. This allows the model to consider more environmental information at the moment when the driver actually makes a decision, rather than just the current environmental information, during the prediction process. By utilizing the seq2seq architecture to model car-following behavior, the model focuses more on reducing the impact of accumulated errors during the training process. Additionally, we propose a temporal consistency constraint loss to improve the robustness and stability of the car-following model’s training method and enhance the prediction results. Finally, we use Gated Recurrent Units (GRU) and the sequence-to-sequence model (seq2seq) as baselines, and simulation results demonstrate that our model achieves more accurate car-following behavior prediction.
- 8 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25 (9) , 12203-12215
A Sequence-to-Sequence Car-Following Model for Addressing Driver Reaction Delay and Cumulative Error in Multi-Step Prediction
Car-following behavior is one of the most common driving behaviors. To reduce the impact of driver reaction delay and accumulated errors in predicting long sequences on the accuracy of speed prediction, we propose a deep learning car-following model based on a sequence-to-sequence (seq2seq) architecture with an attention mechanism. Firstly, we analyze the characteristics of driver reaction delay during the car-following process and design an attention mechanism to learn the probability distribution of driver reaction delay. This allows the model to consider more environmental information at the moment when the driver actually makes a decision, rather than just the current environmental information, during the prediction process. By utilizing the seq2seq architecture to model car-following behavior, the model focuses more on reducing the impact of accumulated errors during the training process. Additionally, we propose a temporal consistency constraint loss to improve the robustness and stability of the car-following model’s training method and enhance the prediction results. Finally, we use Gated Recurrent Units (GRU) and the sequence-to-sequence model (seq2seq) as baselines, and simulation results demonstrate that our model achieves more accurate car-following behavior prediction.
- Open Access
Query-Informed Multi-Agent Motion Prediction
In a dynamic environment, autonomous driving vehicles require accurate decision-making and trajectory planning. To achieve this, autonomous vehicles need to understand their surrounding environment and predict the behavior and future trajectories of other traffic participants. In recent years, vectorization methods have dominated the field of motion prediction due to their ability to capture complex interactions in traffic scenes. However, existing research using vectorization methods for scene encoding often overlooks important physical information about vehicles, such as speed and heading angle, relying solely on displacement to represent the physical attributes of agents. This approach is insufficient for accurate trajectory prediction models. Additionally, agents’ future trajectories can be diverse, such as proceeding straight or making left or right turns at intersections. Therefore, the output of trajectory prediction models should be multimodal to account for these variations. Existing research has used multiple regression heads to output future trajectories and confidence, but the results have been suboptimal. To address these issues, we propose QINET, a method for accurate multimodal trajectory prediction for all agents in a scene. In the scene encoding part, we enhance the feature attributes of agent vehicles to better represent the physical information of agents in the scene. Our scene representation also possesses rotational and spatial invariance. In the decoder part, we use cross-attention and induce the generation of multimodal future trajectories by employing a self-learned query matrix. Experimental results demonstrate that QINET achieves state-of-the-art performance on the Argoverse motion prediction benchmark and is capable of fast multimodal trajectory prediction for multiple agents.
Open AccessQuery-Informed Multi-Agent Motion Prediction
In a dynamic environment, autonomous driving vehicles require accurate decision-making and trajectory planning. To achieve this, autonomous vehicles need to understand their surrounding environment and predict the behavior and future trajectories of other traffic participants. In recent years, vectorization methods have dominated the field of motion prediction due to their ability to capture complex interactions in traffic scenes. However, existing research using vectorization methods for scene encoding often overlooks important physical information about vehicles, such as speed and heading angle, relying solely on displacement to represent the physical attributes of agents. This approach is insufficient for accurate trajectory prediction models. Additionally, agents’ future trajectories can be diverse, such as proceeding straight or making left or right turns at intersections. Therefore, the output of trajectory prediction models should be multimodal to account for these variations. Existing research has used multiple regression heads to output future trajectories and confidence, but the results have been suboptimal. To address these issues, we propose QINET, a method for accurate multimodal trajectory prediction for all agents in a scene. In the scene encoding part, we enhance the feature attributes of agent vehicles to better represent the physical information of agents in the scene. Our scene representation also possesses rotational and spatial invariance. In the decoder part, we use cross-attention and induce the generation of multimodal future trajectories by employing a self-learned query matrix. Experimental results demonstrate that QINET achieves state-of-the-art performance on the Argoverse motion prediction benchmark and is capable of fast multimodal trajectory prediction for multiple agents.
- Open Access
Intelligent Vehicle Decision-Making and Trajectory Planning Method Based on Deep Reinforcement Learning in the Frenet Space
The complexity inherent in navigating intricate traffic environments poses substantial hurdles for intelligent driving technology. The continual progress in mapping and sensor technologies has equipped vehicles with the capability to intricately perceive their exact position and the intricate interplay among surrounding traffic elements. Building upon this foundation, this paper introduces a deep reinforcement learning method to solve the decision-making and trajectory planning problem of intelligent vehicles. The method employs a deep learning framework for feature extraction, utilizing a grid map generated from a blend of static environmental markers such as road centerlines and lane demarcations, in addition to dynamic environmental cues including vehicle positions across varied lanes, all harmonized within the Frenet coordinate system. The grid map serves as the input for the state space, and the input for the action space comprises a vector encompassing lane change timing, velocity, and vertical displacement at the lane change endpoint. To optimize the action strategy, a reinforcement learning approach is employed. The feasibility, stability, and efficiency of the proposed method are substantiated via experiments conducted in the CARLA simulator across diverse driving scenarios, and the proposed method can increase the average success rate of lane change by 6.8% and 13.1% compared with the traditional planning control algorithm and the simple reinforcement learning method.
Open AccessIntelligent Vehicle Decision-Making and Trajectory Planning Method Based on Deep Reinforcement Learning in the Frenet Space
The complexity inherent in navigating intricate traffic environments poses substantial hurdles for intelligent driving technology. The continual progress in mapping and sensor technologies has equipped vehicles with the capability to intricately perceive their exact position and the intricate interplay among surrounding traffic elements. Building upon this foundation, this paper introduces a deep reinforcement learning method to solve the decision-making and trajectory planning problem of intelligent vehicles. The method employs a deep learning framework for feature extraction, utilizing a grid map generated from a blend of static environmental markers such as road centerlines and lane demarcations, in addition to dynamic environmental cues including vehicle positions across varied lanes, all harmonized within the Frenet coordinate system. The grid map serves as the input for the state space, and the input for the action space comprises a vector encompassing lane change timing, velocity, and vertical displacement at the lane change endpoint. To optimize the action strategy, a reinforcement learning approach is employed. The feasibility, stability, and efficiency of the proposed method are substantiated via experiments conducted in the CARLA simulator across diverse driving scenarios, and the proposed method can increase the average success rate of lane change by 6.8% and 13.1% compared with the traditional planning control algorithm and the simple reinforcement learning method.
- 12 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (3) , 3131-3143
Target Track Enhancement Based on Asynchronous Radar and Camera Fusion in Intelligent Driving System
In the domain of intelligent driving vehicles, radar and camera sensors play a crucial role, and their fusion has been widely acknowledged to enhance sensing range and accuracy. This paper delves into the challenge of asynchronous information fusion for object detection between radar and camera, taking into consideration different sampling frequencies of radar and camera. Initially, different state observation and prediction models for radar and camera are formulated, leading to the proposal of a comprehensive target tracking model. Subsequently, the matching and fusion process of detection results is undertaken, yielding an optimal estimation as the output of the sensor information fusion system. Furthermore, the paper introduces the intelligent vehicle sensor fusion task and describes the stepwise approach to sensor fusion in traffic scenarios, including coordinate transform, sensor data preprocessing, and target track. To validate the proposed sensor asynchronous fusion algorithm, a simulation experiment and a road test are conducted focusing on interaction between ego and environment vehicles. The simulation results underscore the effectiveness of the algorithm, with an average lateral error reduction of 32% and an average longitudinal error reduction of 63%. Road test shows that effective fusion can expand the detection range of sensor system and improve the detection performance.
- 12 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (3) , 3131-3143
Target Track Enhancement Based on Asynchronous Radar and Camera Fusion in Intelligent Driving System
In the domain of intelligent driving vehicles, radar and camera sensors play a crucial role, and their fusion has been widely acknowledged to enhance sensing range and accuracy. This paper delves into the challenge of asynchronous information fusion for object detection between radar and camera, taking into consideration different sampling frequencies of radar and camera. Initially, different state observation and prediction models for radar and camera are formulated, leading to the proposal of a comprehensive target tracking model. Subsequently, the matching and fusion process of detection results is undertaken, yielding an optimal estimation as the output of the sensor information fusion system. Furthermore, the paper introduces the intelligent vehicle sensor fusion task and describes the stepwise approach to sensor fusion in traffic scenarios, including coordinate transform, sensor data preprocessing, and target track. To validate the proposed sensor asynchronous fusion algorithm, a simulation experiment and a road test are conducted focusing on interaction between ego and environment vehicles. The simulation results underscore the effectiveness of the algorithm, with an average lateral error reduction of 32% and an average longitudinal error reduction of 63%. Road test shows that effective fusion can expand the detection range of sensor system and improve the detection performance.
- Open Access
- 26 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 121052-121063
A Human-Like Free-Lane-Change Trajectory Planning and Control Method With Data-Based Behavior Decision
Free lane change (FLC) is an important research direction of intelligent driving vehicles. In this paper, a lane change decision model based on deep learning is established with ego and adjacent lane risk and properties of environmental vehicles. A long short-term memory neural network that can be associated with time series characteristics is used to model the lane change process based on analysis of factors affecting the lane change decision. The result of decision-making model training shows that the recognition accuracy of lane change and lane keeping decision reaches more than 92%. Then, a human-like FLC trajectory is planned based on polynomial curve, a quintic polynomial trajectory cluster based on preview distance is generated according to driving conditions, and the optimal FLC trajectory is selected by optimizing an objective function. The model predictive control method is used to dynamically control the vehicle to follow the trajectory. Finally, the lane change decision, trajectory planning, and tracking control model are built in simulation environment, and the control of vehicle dynamics model verifies the integrity and feasibility of free lane change function.
- 26 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 121052-121063
Open AccessA Human-Like Free-Lane-Change Trajectory Planning and Control Method With Data-Based Behavior Decision
Free lane change (FLC) is an important research direction of intelligent driving vehicles. In this paper, a lane change decision model based on deep learning is established with ego and adjacent lane risk and properties of environmental vehicles. A long short-term memory neural network that can be associated with time series characteristics is used to model the lane change process based on analysis of factors affecting the lane change decision. The result of decision-making model training shows that the recognition accuracy of lane change and lane keeping decision reaches more than 92%. Then, a human-like FLC trajectory is planned based on polynomial curve, a quintic polynomial trajectory cluster based on preview distance is generated according to driving conditions, and the optimal FLC trajectory is selected by optimizing an objective function. The model predictive control method is used to dynamically control the vehicle to follow the trajectory. Finally, the lane change decision, trajectory planning, and tracking control model are built in simulation environment, and the control of vehicle dynamics model verifies the integrity and feasibility of free lane change function.
- Open Access
- 7 June 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (11) , 5293
Adaptive Cruise Control for Cut-In Scenarios Based on Model Predictive Control Algorithm
In a cut-in scenario, traditional adaptive cruise control usually cannot effectively identify the cut-in vehicle and respond to it in advance. This paper proposes an adaptive cruise control (ACC) strategy based on the MPC algorithm for cut-in scenarios. A finite state machine (FSM) is designed to manage vehicle control in different cut-in scenarios. For a cut-in scenario, a method to identify and quantify the possibility of cut-in of a vehicle is proposed. At the same time, a safety distance model of the cut-in vehicle is established as the basis for the state transition of the finite state machine. Taking the quantified cut-in possibility of a vehicle as a reference, the model predictive control (MPC) algorithm, which considers the constraints of driving safety and comfort, is used to realize coordinated control of the host vehicle and the cut-in vehicle. Simulink–Carsim simulation studies show that the ACC strategy for a cut-in scenario can effectively identify a cut-in vehicle and quantify the possibility of cut-in of the vehicle. Faced with a cut-in vehicle, the host vehicle using the ACC strategy can brake several seconds early and switch the following target to the cut-in vehicle. Meanwhile, the acceleration and jerk of the host vehicle changes within a reasonable range, which ensures driving safety and comfort.
- 7 June 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (11) , 5293
Open AccessAdaptive Cruise Control for Cut-In Scenarios Based on Model Predictive Control Algorithm
In a cut-in scenario, traditional adaptive cruise control usually cannot effectively identify the cut-in vehicle and respond to it in advance. This paper proposes an adaptive cruise control (ACC) strategy based on the MPC algorithm for cut-in scenarios. A finite state machine (FSM) is designed to manage vehicle control in different cut-in scenarios. For a cut-in scenario, a method to identify and quantify the possibility of cut-in of a vehicle is proposed. At the same time, a safety distance model of the cut-in vehicle is established as the basis for the state transition of the finite state machine. Taking the quantified cut-in possibility of a vehicle as a reference, the model predictive control (MPC) algorithm, which considers the constraints of driving safety and comfort, is used to realize coordinated control of the host vehicle and the cut-in vehicle. Simulink–Carsim simulation studies show that the ACC strategy for a cut-in scenario can effectively identify a cut-in vehicle and quantify the possibility of cut-in of the vehicle. Faced with a cut-in vehicle, the host vehicle using the ACC strategy can brake several seconds early and switch the following target to the cut-in vehicle. Meanwhile, the acceleration and jerk of the host vehicle changes within a reasonable range, which ensures driving safety and comfort.