Locomotor intent classification has become a research hotspot due to its importance to the development of assistive robotics and wearable devices.Previous work have achieved impressive performance in classifying stead...Locomotor intent classification has become a research hotspot due to its importance to the development of assistive robotics and wearable devices.Previous work have achieved impressive performance in classifying steady locomotion states.However,it remains challenging for these methods to attain high accuracy when facing transitions between steady locomotion states.Due to the similarities between the information of the transitions and their adjacent steady states.Furthermore,most of these methods rely solely on data and overlook the objective laws between physical activities,resulting in lower accuracy,particularly when encountering complex locomotion modes such as transitions.To address the existing deficiencies,we propose the locomotion rule embedding long short-term memory(LSTM)network with Attention(LREAL)for human locomotor intent classification,with a particular focus on transitions,using data from fewer sensors(two inertial measurement units and four goniometers).The LREAL network consists of two levels:One responsible for distinguishing between steady states and transitions,and the other for the accurate identification of locomotor intent.Each classifier in these levels is composed of multiple-LSTM layers and an attention mechanism.To introduce real-world motion rules and apply constraints to the network,a prior knowledge was added to the network via a rule-modulating block.The method was tested on the ENABL3S dataset,which contains continuous locomotion date for seven steady and twelve transitions states.Experimental results showed that the LREAL network could recognize locomotor intents with an average accuracy of 99.03%and 96.52%for the steady and transitions states,respectively.It is worth noting that the LREAL network accuracy for transition-state recognition improved by 0.18%compared to other state-of-the-art network,while using data from fewer sensors.展开更多
Although previous studies have made some clear leap in learning latent dynamics from high-dimensional representations,the performances in terms of accuracy and inference time of long-term model prediction still need t...Although previous studies have made some clear leap in learning latent dynamics from high-dimensional representations,the performances in terms of accuracy and inference time of long-term model prediction still need to be improved.In this study,a deep convolutional network based on the Koopman operator(CKNet)is proposed to model non-linear systems with pixel-level measurements for long-term prediction.CKNet adopts an autoencoder network architecture,consisting of an encoder to generate latent states and a linear dynamical model(i.e.,the Koopman operator)which evolves in the latent state space spanned by the encoder.The decoder is used to recover images from latent states.According to a multi-step ahead prediction loss function,the system matrices for approximating the Koopman operator are trained synchronously with the autoencoder in a mini-batch manner.In this manner,gradients can be synchronously transmitted to both the system matrices and the autoencoder to help the encoder self-adaptively tune the latent state space in the training process,and the resulting model is time-invariant in the latent space.Therefore,the proposed CKNet has the advantages of less inference time and high accuracy for long-term prediction.Experiments are per-formed on OpenAI Gym and Mujoco environments,including two and four non-linear forced dynamical systems with continuous action spaces.The experimental results show that CKNet has strong long-term prediction capabilities with sufficient precision.展开更多
The 3D reconstruction using deep learning-based intelligent systems can provide great help for measuring an individual’s height and shape quickly and accurately through 2D motion-blurred images.Generally,during the a...The 3D reconstruction using deep learning-based intelligent systems can provide great help for measuring an individual’s height and shape quickly and accurately through 2D motion-blurred images.Generally,during the acquisition of images in real-time,motion blur,caused by camera shaking or human motion,appears.Deep learning-based intelligent control applied in vision can help us solve the problem.To this end,we propose a 3D reconstruction method for motion-blurred images using deep learning.First,we develop a BF-WGAN algorithm that combines the bilateral filtering(BF)denoising theory with a Wasserstein generative adversarial network(WGAN)to remove motion blur.The bilateral filter denoising algorithm is used to remove the noise and to retain the details of the blurred image.Then,the blurred image and the corresponding sharp image are input into the WGAN.This algorithm distinguishes the motion-blurred image from the corresponding sharp image according to the WGAN loss and perceptual loss functions.Next,we use the deblurred images generated by the BFWGAN algorithm for 3D reconstruction.We propose a threshold optimization random sample consensus(TO-RANSAC)algorithm that can remove the wrong relationship between two views in the 3D reconstructed model relatively accurately.Compared with the traditional RANSAC algorithm,the TO-RANSAC algorithm can adjust the threshold adaptively,which improves the accuracy of the 3D reconstruction results.The experimental results show that our BF-WGAN algorithm has a better deblurring effect and higher efficiency than do other representative algorithms.In addition,the TO-RANSAC algorithm yields a calculation accuracy considerably higher than that of the traditional RANSAC algorithm.展开更多
In this study,we propose a deep-learning-based method to correct motion artifacts in optical resolution photoacoustic microscopy(OR-PAM).The method is a convolutional neural network that establishes an end-to-end map ...In this study,we propose a deep-learning-based method to correct motion artifacts in optical resolution photoacoustic microscopy(OR-PAM).The method is a convolutional neural network that establishes an end-to-end map from input raw data with motion artifacts to output corrected images.First,we performed simulation studies to evaluate the feasibility and effectiveness of the proposed method.Second,we employed this method to process images of rat brain vessels with multiple motion artifacts to evaluate its performance for in vivo applications.The results demonstrate that this method works well for both large blood vessels and capillary networks.In comparison with traditional methods,the proposed method in this study can be easily modified to satisfy different scenarios of motion corrections in OR-PAM by revising the training sets.展开更多
A Deep Draft Semi-submersible (DDS) under certain flow conditions could be subjected to Vortex-Induced Motions (VIM), which significantly influences the loads on and life fatigue of the moorings and the risers. To...A Deep Draft Semi-submersible (DDS) under certain flow conditions could be subjected to Vortex-Induced Motions (VIM), which significantly influences the loads on and life fatigue of the moorings and the risers. To investigate the VIM of a DDS with four rectangular section columns in waves coupled with a uniform current, a numerical study using the computational fluid dynamics (CFD) method was conducted. The issues of the VIM of multi-column floaters can be con','eniently converted to the issues of oscillating cylinders in fluid cross flows. This paper looks into the CFD numerical simulation of infinite cylinders having rectangular sections in a two-dimensional sinusoidal time- dependent flow field coupled with a uniform current. The resulted hydrodynamic forces and motion responses in different oscillatory flows plus currents both aligned in the same direction for the incidence of 135° of the DDS relative to the flow are compared with the ones in current only cases. The results show that the VIM response of this geometric arrangement of a DDS with four rectangular columns in a current combined with oscillatory flows is more evident than that in the current only case. The oscillatory flows and waves have the significant influence on the VIM response, forces and trajectory, in-plane motions of the DDS.展开更多
Dynamic failure of rock masses around deep tunnels,such as fault-slip rockburst and seismic-induced collapse,can pose a significant threat to tunnel construction safety.One of the most significant factors that control...Dynamic failure of rock masses around deep tunnels,such as fault-slip rockburst and seismic-induced collapse,can pose a significant threat to tunnel construction safety.One of the most significant factors that control the accuracy of its risk assessment is the estimation of the ground motion around a tunnel caused by seismicity events.In general,the characteristic parameters of ground motion are estimated in terms of empirical scaling laws.However,these scaling laws make it difficult to accurately estimate the near-field ground motion parameters because the roles of control factors,such as tunnel geometry,damage zone distribution,and seismic source parameters,are not considered.For this,the finite fracturing seismic source model(FFSSM)proposed in this study is used to simulate the near-field ground motion characteristics around deep tunnels.Then,the amplification effects of ground motion caused by the interaction between seismic waves and deep tunnels and corresponding control factors are studied.The control effects of four factors on the near-field ground motion amplification effect are analyzed,including the main seismic source wavelength,tunnel span,tunnel shape,and range of damage zones.An empirical formula for the maximum amplification factor(a_(m))of the near-field ground motion around deep tunnels is proposed,which consists of four control factors,i.e.the wavelength control factor(F_(λ)),tunnel span factor(F_(D)),tunnel shape factor(F_(s))and excavation damage factor(F_(d)).This empirical formula provides an easy approach for accurately estimating the ground motion parameters in seismicityprone regimes and the rock support design of deep tunnels under dynamic loads.展开更多
Mooring arrays have been widely deployed in sustained ocean observation in high resolution to measure finer dynamic features of marine phenomena.However,the irregular posture changes and nonlinear response of moorings...Mooring arrays have been widely deployed in sustained ocean observation in high resolution to measure finer dynamic features of marine phenomena.However,the irregular posture changes and nonlinear response of moorings under the effect of ocean currents face huge challenges for the deployment of mooring arrays,which may cause the deviations of measurements and yield a vacuum of observation in the upper ocean.We developed a data-driven mooring simulation model based on LSTM(long short-term memory)neural network,coupling the ocean current with position data from moorings to predict the motion of moorings,including single-step output prediction and multi-step prediction.Based on the predictive information,the formation of the mooring array can be adjusted to improve the accuracy and integrity of measurements.Moreover,we proposed the cuckoo search(CS)optimization algorithm to tune the parameters of LSTM,which improves the robustness and generalization of the model.We utilize the datasets observed from moorings anchored in the Kuroshio Extension region to train and validate the simulation model.The experimental results demonstrate that the model can remarkably improve prediction accuracy and yield stable performance.Moreover,compared with other optimization algorithms,CS is more efficient and performs better in simulating the motion of moorings.展开更多
4-Dimensional cone-beam computed tomography(4D-CBCT)offers several key advantages over conventional 3DCBCT in moving target localization/delineation,structure de-blurring,target motion tracking,treatment dose accumul...4-Dimensional cone-beam computed tomography(4D-CBCT)offers several key advantages over conventional 3DCBCT in moving target localization/delineation,structure de-blurring,target motion tracking,treatment dose accumulation and adaptive radiation therapy.However,the use of the 4D-CBCT in current radiation therapy practices has been limited,mostly due to its sub-optimal image quality from limited angular sampling of conebeam projections.In this study,we summarized the recent developments of 4D-CBCT reconstruction techniques for image quality improvement,and introduced our developments of a new 4D-CBCT reconstruction technique which features simultaneous motion estimation and image reconstruction(SMEIR).Based on the original SMEIR scheme,biomechanical modeling-guided SMEIR(SMEIR-Bio)was introduced to further improve the reconstruction accuracy of fine details in lung 4D-CBCTs.To improve the efficiency of reconstruction,we recently developed a U-net-based deformation-vector-field(DVF)optimization technique to leverage a population-based deep learning scheme to improve the accuracy of intra-lung DVFs(SMEIR-Unet),without explicit biomechanical modeling.Details of each of the SMEIR,SMEIR-Bio and SMEIR-Unet techniques were included in this study,along with the corresponding results comparing the reconstruction accuracy in terms of CBCT images and the DVFs.We also discussed the application prospects of the SMEIR-type techniques in image-guided radiation therapy and adaptive radiation therapy,and presented potential schemes on future developments to achieve faster and more accurate 4D-CBCT imaging.展开更多
Brain-computer interfaces(BCI)use neural activity as a control signal to enable direct communication between the human brain and external devices.The electrical signals generated by the brain are captured through elec...Brain-computer interfaces(BCI)use neural activity as a control signal to enable direct communication between the human brain and external devices.The electrical signals generated by the brain are captured through electroencephalogram(EEG)and translated into neural intentions reflecting the user’s behavior.Correct decoding of the neural intentions then facilitates the control of external devices.Reinforcement learning-based BCIs enhance decoders to complete tasks based only on feedback signals(rewards)from the environment,building a general framework for dynamic mapping from neural intentions to actions that adapt to changing environments.However,using traditional reinforcement learning methods can have challenges such as the curse of dimensionality and poor generalization.Therefore,in this paper,we use deep reinforcement learning to construct decoders for the correct decoding of EEG signals,demonstrate its feasibility through experiments,and demonstrate its stronger generalization on motion imaging(MI)EEG data signals with high dynamic characteristics.展开更多
Decision-making and motion planning are extremely important in autonomous driving to ensure safe driving in a real-world environment.This study proposes an online evolutionary decision-making and motion planning frame...Decision-making and motion planning are extremely important in autonomous driving to ensure safe driving in a real-world environment.This study proposes an online evolutionary decision-making and motion planning framework for autonomous driving based on a hybrid data-and model-driven method.First,a data-driven decision-making module based on deep reinforcement learning(DRL)is developed to pursue a rational driving performance as much as possible.Then,model predictive control(MPC)is employed to execute both longitudinal and lateral motion planning tasks.Multiple constraints are defined according to the vehicle’s physical limit to meet the driving task requirements.Finally,two principles of safety and rationality for the self-evolution of autonomous driving are proposed.A motion envelope is established and embedded into a rational exploration and exploitation scheme,which filters out unreasonable experiences by masking unsafe actions so as to collect high-quality training data for the DRL agent.Experiments with a high-fidelity vehicle model and MATLAB/Simulink co-simulation environment are conducted,and the results show that the proposed online-evolution framework is able to generate safer,more rational,and more efficient driving action in a real-world environment.展开更多
基金funded by the National Natural Science Foundation of China(Nos.62072212,62302218)the Development Project of Jilin Province of China(Nos.20220508125RC,20230201065GX,20240101364JC)+1 种基金National Key R&D Program(No.2018YFC2001302)the Jilin Provincial Key Laboratory of Big Data Intelligent Cognition(No.20210504003GH).
文摘Locomotor intent classification has become a research hotspot due to its importance to the development of assistive robotics and wearable devices.Previous work have achieved impressive performance in classifying steady locomotion states.However,it remains challenging for these methods to attain high accuracy when facing transitions between steady locomotion states.Due to the similarities between the information of the transitions and their adjacent steady states.Furthermore,most of these methods rely solely on data and overlook the objective laws between physical activities,resulting in lower accuracy,particularly when encountering complex locomotion modes such as transitions.To address the existing deficiencies,we propose the locomotion rule embedding long short-term memory(LSTM)network with Attention(LREAL)for human locomotor intent classification,with a particular focus on transitions,using data from fewer sensors(two inertial measurement units and four goniometers).The LREAL network consists of two levels:One responsible for distinguishing between steady states and transitions,and the other for the accurate identification of locomotor intent.Each classifier in these levels is composed of multiple-LSTM layers and an attention mechanism.To introduce real-world motion rules and apply constraints to the network,a prior knowledge was added to the network via a rule-modulating block.The method was tested on the ENABL3S dataset,which contains continuous locomotion date for seven steady and twelve transitions states.Experimental results showed that the LREAL network could recognize locomotor intents with an average accuracy of 99.03%and 96.52%for the steady and transitions states,respectively.It is worth noting that the LREAL network accuracy for transition-state recognition improved by 0.18%compared to other state-of-the-art network,while using data from fewer sensors.
基金National Natural Science Foundation of China,Grant/Award Numbers:61825305,62003361,U21A20518China Postdoctoral Science Foundation,Grant/Award Number:47680。
文摘Although previous studies have made some clear leap in learning latent dynamics from high-dimensional representations,the performances in terms of accuracy and inference time of long-term model prediction still need to be improved.In this study,a deep convolutional network based on the Koopman operator(CKNet)is proposed to model non-linear systems with pixel-level measurements for long-term prediction.CKNet adopts an autoencoder network architecture,consisting of an encoder to generate latent states and a linear dynamical model(i.e.,the Koopman operator)which evolves in the latent state space spanned by the encoder.The decoder is used to recover images from latent states.According to a multi-step ahead prediction loss function,the system matrices for approximating the Koopman operator are trained synchronously with the autoencoder in a mini-batch manner.In this manner,gradients can be synchronously transmitted to both the system matrices and the autoencoder to help the encoder self-adaptively tune the latent state space in the training process,and the resulting model is time-invariant in the latent space.Therefore,the proposed CKNet has the advantages of less inference time and high accuracy for long-term prediction.Experiments are per-formed on OpenAI Gym and Mujoco environments,including two and four non-linear forced dynamical systems with continuous action spaces.The experimental results show that CKNet has strong long-term prediction capabilities with sufficient precision.
基金the National Natural Science Foundation of China under Grant 61902311in part by the Japan Society for the Promotion of Science(JSPS)Grants-in-Aid for Scientific Research(KAKENHI)under Grant JP18K18044.
文摘The 3D reconstruction using deep learning-based intelligent systems can provide great help for measuring an individual’s height and shape quickly and accurately through 2D motion-blurred images.Generally,during the acquisition of images in real-time,motion blur,caused by camera shaking or human motion,appears.Deep learning-based intelligent control applied in vision can help us solve the problem.To this end,we propose a 3D reconstruction method for motion-blurred images using deep learning.First,we develop a BF-WGAN algorithm that combines the bilateral filtering(BF)denoising theory with a Wasserstein generative adversarial network(WGAN)to remove motion blur.The bilateral filter denoising algorithm is used to remove the noise and to retain the details of the blurred image.Then,the blurred image and the corresponding sharp image are input into the WGAN.This algorithm distinguishes the motion-blurred image from the corresponding sharp image according to the WGAN loss and perceptual loss functions.Next,we use the deblurred images generated by the BFWGAN algorithm for 3D reconstruction.We propose a threshold optimization random sample consensus(TO-RANSAC)algorithm that can remove the wrong relationship between two views in the 3D reconstructed model relatively accurately.Compared with the traditional RANSAC algorithm,the TO-RANSAC algorithm can adjust the threshold adaptively,which improves the accuracy of the 3D reconstruction results.The experimental results show that our BF-WGAN algorithm has a better deblurring effect and higher efficiency than do other representative algorithms.In addition,the TO-RANSAC algorithm yields a calculation accuracy considerably higher than that of the traditional RANSAC algorithm.
基金This work was sponsored by National Natural Science Foundation of China,Nos.81571722,61775028 and 61528401.
文摘In this study,we propose a deep-learning-based method to correct motion artifacts in optical resolution photoacoustic microscopy(OR-PAM).The method is a convolutional neural network that establishes an end-to-end map from input raw data with motion artifacts to output corrected images.First,we performed simulation studies to evaluate the feasibility and effectiveness of the proposed method.Second,we employed this method to process images of rat brain vessels with multiple motion artifacts to evaluate its performance for in vivo applications.The results demonstrate that this method works well for both large blood vessels and capillary networks.In comparison with traditional methods,the proposed method in this study can be easily modified to satisfy different scenarios of motion corrections in OR-PAM by revising the training sets.
基金supported by the National Natural Science Foundation of China(Grant No.51279104)a Research Project on High-Technology Ships by the Ministry of Industry and Information Technology
文摘A Deep Draft Semi-submersible (DDS) under certain flow conditions could be subjected to Vortex-Induced Motions (VIM), which significantly influences the loads on and life fatigue of the moorings and the risers. To investigate the VIM of a DDS with four rectangular section columns in waves coupled with a uniform current, a numerical study using the computational fluid dynamics (CFD) method was conducted. The issues of the VIM of multi-column floaters can be con','eniently converted to the issues of oscillating cylinders in fluid cross flows. This paper looks into the CFD numerical simulation of infinite cylinders having rectangular sections in a two-dimensional sinusoidal time- dependent flow field coupled with a uniform current. The resulted hydrodynamic forces and motion responses in different oscillatory flows plus currents both aligned in the same direction for the incidence of 135° of the DDS relative to the flow are compared with the ones in current only cases. The results show that the VIM response of this geometric arrangement of a DDS with four rectangular columns in a current combined with oscillatory flows is more evident than that in the current only case. The oscillatory flows and waves have the significant influence on the VIM response, forces and trajectory, in-plane motions of the DDS.
基金jointly supported by the National Natural Science Foundation of China(Grant No.41877256)the Natural Science Foundation of Hubei Province(Grant No.ZRQT2020000114)the Key Research Program of the Chinese Academy of Sciences(Grant No.KFZD-SW-423)。
文摘Dynamic failure of rock masses around deep tunnels,such as fault-slip rockburst and seismic-induced collapse,can pose a significant threat to tunnel construction safety.One of the most significant factors that control the accuracy of its risk assessment is the estimation of the ground motion around a tunnel caused by seismicity events.In general,the characteristic parameters of ground motion are estimated in terms of empirical scaling laws.However,these scaling laws make it difficult to accurately estimate the near-field ground motion parameters because the roles of control factors,such as tunnel geometry,damage zone distribution,and seismic source parameters,are not considered.For this,the finite fracturing seismic source model(FFSSM)proposed in this study is used to simulate the near-field ground motion characteristics around deep tunnels.Then,the amplification effects of ground motion caused by the interaction between seismic waves and deep tunnels and corresponding control factors are studied.The control effects of four factors on the near-field ground motion amplification effect are analyzed,including the main seismic source wavelength,tunnel span,tunnel shape,and range of damage zones.An empirical formula for the maximum amplification factor(a_(m))of the near-field ground motion around deep tunnels is proposed,which consists of four control factors,i.e.the wavelength control factor(F_(λ)),tunnel span factor(F_(D)),tunnel shape factor(F_(s))and excavation damage factor(F_(d)).This empirical formula provides an easy approach for accurately estimating the ground motion parameters in seismicityprone regimes and the rock support design of deep tunnels under dynamic loads.
基金Supported by the Laoshan Laboratory (Nos.LSKJ202201302-5,LSKJ202201405-1,LSKJ202204304)。
文摘Mooring arrays have been widely deployed in sustained ocean observation in high resolution to measure finer dynamic features of marine phenomena.However,the irregular posture changes and nonlinear response of moorings under the effect of ocean currents face huge challenges for the deployment of mooring arrays,which may cause the deviations of measurements and yield a vacuum of observation in the upper ocean.We developed a data-driven mooring simulation model based on LSTM(long short-term memory)neural network,coupling the ocean current with position data from moorings to predict the motion of moorings,including single-step output prediction and multi-step prediction.Based on the predictive information,the formation of the mooring array can be adjusted to improve the accuracy and integrity of measurements.Moreover,we proposed the cuckoo search(CS)optimization algorithm to tune the parameters of LSTM,which improves the robustness and generalization of the model.We utilize the datasets observed from moorings anchored in the Kuroshio Extension region to train and validate the simulation model.The experimental results demonstrate that the model can remarkably improve prediction accuracy and yield stable performance.Moreover,compared with other optimization algorithms,CS is more efficient and performs better in simulating the motion of moorings.
基金This work was supported in part by grants from the US National Institutes of Health,Nos.R01 EB020366 and R01 EB027898the Cancer Prevention and Research Institute of Texas,Nos.RP130109 and RP160661from the University of Texas Southwestern Medical Center(Radiation Oncology Seed Grant).
文摘4-Dimensional cone-beam computed tomography(4D-CBCT)offers several key advantages over conventional 3DCBCT in moving target localization/delineation,structure de-blurring,target motion tracking,treatment dose accumulation and adaptive radiation therapy.However,the use of the 4D-CBCT in current radiation therapy practices has been limited,mostly due to its sub-optimal image quality from limited angular sampling of conebeam projections.In this study,we summarized the recent developments of 4D-CBCT reconstruction techniques for image quality improvement,and introduced our developments of a new 4D-CBCT reconstruction technique which features simultaneous motion estimation and image reconstruction(SMEIR).Based on the original SMEIR scheme,biomechanical modeling-guided SMEIR(SMEIR-Bio)was introduced to further improve the reconstruction accuracy of fine details in lung 4D-CBCTs.To improve the efficiency of reconstruction,we recently developed a U-net-based deformation-vector-field(DVF)optimization technique to leverage a population-based deep learning scheme to improve the accuracy of intra-lung DVFs(SMEIR-Unet),without explicit biomechanical modeling.Details of each of the SMEIR,SMEIR-Bio and SMEIR-Unet techniques were included in this study,along with the corresponding results comparing the reconstruction accuracy in terms of CBCT images and the DVFs.We also discussed the application prospects of the SMEIR-type techniques in image-guided radiation therapy and adaptive radiation therapy,and presented potential schemes on future developments to achieve faster and more accurate 4D-CBCT imaging.
文摘Brain-computer interfaces(BCI)use neural activity as a control signal to enable direct communication between the human brain and external devices.The electrical signals generated by the brain are captured through electroencephalogram(EEG)and translated into neural intentions reflecting the user’s behavior.Correct decoding of the neural intentions then facilitates the control of external devices.Reinforcement learning-based BCIs enhance decoders to complete tasks based only on feedback signals(rewards)from the environment,building a general framework for dynamic mapping from neural intentions to actions that adapt to changing environments.However,using traditional reinforcement learning methods can have challenges such as the curse of dimensionality and poor generalization.Therefore,in this paper,we use deep reinforcement learning to construct decoders for the correct decoding of EEG signals,demonstrate its feasibility through experiments,and demonstrate its stronger generalization on motion imaging(MI)EEG data signals with high dynamic characteristics.
基金the financial support of the National Key Research and Development Program of China(2020AAA0108100)the Shanghai Municipal Science and Technology Major Project(2021SHZDZX0100)the Shanghai Gaofeng and Gaoyuan Project for University Academic Program Development for funding。
文摘Decision-making and motion planning are extremely important in autonomous driving to ensure safe driving in a real-world environment.This study proposes an online evolutionary decision-making and motion planning framework for autonomous driving based on a hybrid data-and model-driven method.First,a data-driven decision-making module based on deep reinforcement learning(DRL)is developed to pursue a rational driving performance as much as possible.Then,model predictive control(MPC)is employed to execute both longitudinal and lateral motion planning tasks.Multiple constraints are defined according to the vehicle’s physical limit to meet the driving task requirements.Finally,two principles of safety and rationality for the self-evolution of autonomous driving are proposed.A motion envelope is established and embedded into a rational exploration and exploitation scheme,which filters out unreasonable experiences by masking unsafe actions so as to collect high-quality training data for the DRL agent.Experiments with a high-fidelity vehicle model and MATLAB/Simulink co-simulation environment are conducted,and the results show that the proposed online-evolution framework is able to generate safer,more rational,and more efficient driving action in a real-world environment.