-
公开(公告)号：US20240038066A1
公开(公告)日：2024-02-01
申请号：US18026835
申请日：2022-08-04
Applicant: JIANGSU UNIVERSITY
Inventor： Yingfeng CAI , Sikai LU , Long CHEN , Hai WANG , Chaochun YUAN , Qingchao LIU , Yicheng LI
IPC: G08G1/08
CPC classification number: G08G1/08
Abstract: A multi-intelligence federated reinforcement learning (FRL)-based vehicle-road cooperative control system and method at the complex intersection use a vehicle-road cooperative control framework based on the Road Side Unit (RSU) static processing module and the vehicle-based dynamic processing module. The historical road information is supplied by the proposed RSU module. The Federated Twin Delayed Deep Deterministic policy gradient (FTD3) algorithm is proposed to connect the federated learning (FL) module and the reinforcement learning (RL) module. The FTD3 algorithm transmits only neural network parameters instead of vehicle samples to protect privacy. Firstly, FTD3 selects only specific networks for aggregation to reduce the communication cost. Secondly, FTD3 realizes the deep combination of FL and RL by aggregating target critic networks with smaller Q-values. Thirdly, RSU neural network participates in aggregation rather than training, and only shared global model parameters are used.
-
2.发明申请公开(公告)号：US20240420487A1
公开(公告)日：2024-12-19
申请号：US18575391
申请日：2023-05-06
Applicant: JIANGSU UNIVERSITY
Inventor： Hai WANG , Guirong ZHANG , Yingfeng CAI , Long CHEN , Yicheng LI , Qingchao LIU
Abstract: A multi-task joint perception network model and detection method for traffic road surface information can simultaneously detect a lane line and a drivable area. A coordinate attention mechanism is integrated into a traditional feature extraction network to ensure that a feature extraction effect is enhanced while a calculated amount is not increased. In a neck network, a dilated convolution residual module is proposed to enhance performance of prediction of details by the network, and a decoder part shares features of the drivable area into lane line detection to enhance a lane line detection effect under complex road conditions. In a training stage, there is provided a alternating optimization training method to improve integral segmentation performance of the model. The multi-task joint perception network model and detection method realizes quite high accuracy and excellent speed performance in a challenging BDD100K dataset.
-
公开(公告)号：US20210276548A1
公开(公告)日：2021-09-09
申请号：US16626886
申请日：2019-02-20
Applicant: JIANGSU UNIVERSITY
Inventor： Yingfeng CAI , Yong ZANG , Hai WANG , Xiaoqiang SUN , Long CHEN , Jun LIANG , Yicheng LI , Dehua SHI , Bin TANG
Abstract: This invention is an extension adaptive lane keeping control method with variable vehicle speed, which is composed of the following steps: S1, establishing a three-degree-of-freedom dynamic model and a preview deviation expression; S2, performing the lane line fitting equation; S3, designing the upper layer ISTE extension controller; including: S3.1, establishing the control index (ISTE) extension sets; S3.2, dividing the control index (ISTE) domain boundaries; S3.3, calculating the control index (ISTE) association function; S3.4, establishing the upper layer extension controller decision; S4, designing the lower layer speed extension controller; S5, designing the lower layer deviation tracking extension controller; including: S5.1, extracting the lower layer deviation tracking extension feature quantity and dividing domain boundaries; S5.2, designing the lower layer extension controller correlation function; S5.3, performing the lower layer measurement mode identification; S5.4, When the front wheel angle of lower layer controller outputs is calculated according to the measurement mode. This invention realizes the adaptive variation of the control coefficient of the extension controller and the boundary range of the constraint domain according to the tracking deviation precision, the speed variation, and the expert knowledge base.
-
-
Search Results
3
records
(took 0.009 seconds)
