To reduce the discrepancy between the source and target domains,a new multi-label adaptation network(ML-ANet)based on multiple kernel variants with maximum mean discrepancies is proposed in this paper.The hidden repre...To reduce the discrepancy between the source and target domains,a new multi-label adaptation network(ML-ANet)based on multiple kernel variants with maximum mean discrepancies is proposed in this paper.The hidden representations of the task-specific layers in ML-ANet are embedded in the reproducing kernel Hilbert space(RKHS)so that the mean-embeddings of specific features in different domains could be precisely matched.Multiple kernel functions are used to improve feature distribution efficiency for explicit mean embedding matching,which can further reduce domain discrepancy.Adverse weather and cross-camera adaptation examinations are conducted to verify the effectiveness of our proposed ML-ANet.The results show that our proposed ML-ANet achieves higher accuracies than the compared state-of-the-art methods for multi-label image classification in both the adverse weather adaptation and cross-camera adaptation experiments.These results indicate that ML-ANet can alleviate the reliance on fully labeled training data and improve the accuracy of multi-label image classification in various domain shift scenarios.展开更多
Road intersection is one of the most complex and accident-prone traffic scenarios,so it’s challenging for autonomous vehicles(AVs)to make safe and efficient decisions at the intersections.Most of the related studies ...Road intersection is one of the most complex and accident-prone traffic scenarios,so it’s challenging for autonomous vehicles(AVs)to make safe and efficient decisions at the intersections.Most of the related studies focus on the solution to a single scenario or only guarantee safety without considering driving efficiency.To address these problems,this study proposed a deep reinforcement learning enabled decision-making framework for AVs to drive through intersections automatically,safely and efficiently.The mapping relationship between traffic images and vehicle operations was obtained by an end-to-end decision-making framework established by convolutional neural networks.Traffic images collected at two timesteps were used to calculate the relative velocity between vehicles.Markov decision process was employed to model the interaction between AVs and other vehicles,and the deep Q-network algorithm was utilized to obtain the optimal driving policy regarding safety and efficiency.To verify the effectiveness of the proposed decision-making framework,the top three accident-prone crossing path crash scenarios at intersections were simulated,when different initial vehicle states were adopted for better generalization capability.The results showed that the developed method could make AVs drive safely and efficiently through intersections in all of the tested scenarios.展开更多
This paper proposes an improved decision-making method based on deep reinforcement learning to address on-ramp merging challenges in highway autonomous driving.A novel safety indicator,time difference to merging(TDTM)...This paper proposes an improved decision-making method based on deep reinforcement learning to address on-ramp merging challenges in highway autonomous driving.A novel safety indicator,time difference to merging(TDTM),is introduced,which is used in conjunction with the classic time to collision(TTC)indicator to evaluate driving safety and assist the merging vehicle in finding a suitable gap in traffic,thereby enhancing driving safety.The training of an autonomous driving agent is performed using the Deep Deterministic Policy Gradient(DDPG)algorithm.An action-masking mechanism is deployed to prevent unsafe actions during the policy exploration phase.The proposed DDPG+TDTM+TTC solution is tested in on-ramp merging scenarios with different driving speeds in SUMO and achieves a success rate of 99.96%without significantly impacting traffic efficiency on the main road.The results demonstrate that DDPG+TDTM+TTC achieved a higher on-ramp merging success rate of 99.96%compared to DDPG+TTC and DDPG.展开更多
Estimating depth from images captured by camera sensors is crucial for the advancement of autonomous driving technologies and has gained significant attention in recent years.However,most previous methods rely on stac...Estimating depth from images captured by camera sensors is crucial for the advancement of autonomous driving technologies and has gained significant attention in recent years.However,most previous methods rely on stacked pooling or stride convolution to extract high-level features,which can limit network performance and lead to information redundancy.This paper proposes an improved bidirectional feature pyramid module(BiFPN)and a channel attention module(Seblock:squeeze and excitation)to address these issues in existing methods based on monocular camera sensor.The Seblock redistributes channel feature weights to enhance useful information,while the improved BiFPN facilitates efficient fusion of multi-scale features.The proposed method is in an end-to-end solution without any additional post-processing,resulting in efficient depth estimation.Experiment results show that the proposed method is competitive with state-of-the-art algorithms and preserves fine-grained texture of scene depth.展开更多
基金Supported by Shenzhen Fundamental Research Fund of China(Grant No.JCYJ20190808142613246)National Natural Science Foundation of China(Grant No.51805332),and Young Elite Scientists Sponsorship Program funded by the China Society of Automotive Engineers.
文摘To reduce the discrepancy between the source and target domains,a new multi-label adaptation network(ML-ANet)based on multiple kernel variants with maximum mean discrepancies is proposed in this paper.The hidden representations of the task-specific layers in ML-ANet are embedded in the reproducing kernel Hilbert space(RKHS)so that the mean-embeddings of specific features in different domains could be precisely matched.Multiple kernel functions are used to improve feature distribution efficiency for explicit mean embedding matching,which can further reduce domain discrepancy.Adverse weather and cross-camera adaptation examinations are conducted to verify the effectiveness of our proposed ML-ANet.The results show that our proposed ML-ANet achieves higher accuracies than the compared state-of-the-art methods for multi-label image classification in both the adverse weather adaptation and cross-camera adaptation experiments.These results indicate that ML-ANet can alleviate the reliance on fully labeled training data and improve the accuracy of multi-label image classification in various domain shift scenarios.
基金This work is supported by the National Natural Science Foundation of China(Grant No.51805332)the Young Elite Scientists Sponsorship Program funded by the China Society of Automotive Engineers,the Natural Science Foundation of Guangdong Province(Grant No.2018A030310532)the Shenzhen Fundamental Research Fund(Grant No.JCYJ20190808142613246).
文摘Road intersection is one of the most complex and accident-prone traffic scenarios,so it’s challenging for autonomous vehicles(AVs)to make safe and efficient decisions at the intersections.Most of the related studies focus on the solution to a single scenario or only guarantee safety without considering driving efficiency.To address these problems,this study proposed a deep reinforcement learning enabled decision-making framework for AVs to drive through intersections automatically,safely and efficiently.The mapping relationship between traffic images and vehicle operations was obtained by an end-to-end decision-making framework established by convolutional neural networks.Traffic images collected at two timesteps were used to calculate the relative velocity between vehicles.Markov decision process was employed to model the interaction between AVs and other vehicles,and the deep Q-network algorithm was utilized to obtain the optimal driving policy regarding safety and efficiency.To verify the effectiveness of the proposed decision-making framework,the top three accident-prone crossing path crash scenarios at intersections were simulated,when different initial vehicle states were adopted for better generalization capability.The results showed that the developed method could make AVs drive safely and efficiently through intersections in all of the tested scenarios.
基金supported by the National Natural Science Foundation of China(Grant No.52272421)the Shenzhen Fundamental Research Fund(Grant No.JCYJ20190808142613246).
文摘This paper proposes an improved decision-making method based on deep reinforcement learning to address on-ramp merging challenges in highway autonomous driving.A novel safety indicator,time difference to merging(TDTM),is introduced,which is used in conjunction with the classic time to collision(TTC)indicator to evaluate driving safety and assist the merging vehicle in finding a suitable gap in traffic,thereby enhancing driving safety.The training of an autonomous driving agent is performed using the Deep Deterministic Policy Gradient(DDPG)algorithm.An action-masking mechanism is deployed to prevent unsafe actions during the policy exploration phase.The proposed DDPG+TDTM+TTC solution is tested in on-ramp merging scenarios with different driving speeds in SUMO and achieves a success rate of 99.96%without significantly impacting traffic efficiency on the main road.The results demonstrate that DDPG+TDTM+TTC achieved a higher on-ramp merging success rate of 99.96%compared to DDPG+TTC and DDPG.
基金supported by the National Natural Science Foundation of China(Grant No.52272421)Shenzhen Fundamental Research Fund(Grant Number:JCYJ20190808142613246 and 20200803015912001).
文摘Estimating depth from images captured by camera sensors is crucial for the advancement of autonomous driving technologies and has gained significant attention in recent years.However,most previous methods rely on stacked pooling or stride convolution to extract high-level features,which can limit network performance and lead to information redundancy.This paper proposes an improved bidirectional feature pyramid module(BiFPN)and a channel attention module(Seblock:squeeze and excitation)to address these issues in existing methods based on monocular camera sensor.The Seblock redistributes channel feature weights to enhance useful information,while the improved BiFPN facilitates efficient fusion of multi-scale features.The proposed method is in an end-to-end solution without any additional post-processing,resulting in efficient depth estimation.Experiment results show that the proposed method is competitive with state-of-the-art algorithms and preserves fine-grained texture of scene depth.
