Pub Date : 2022-12-05DOI: 10.1109/ROBIO55434.2022.10011945
Wei-qing Xu, Yue Gao, Buqing Nie
Recently, Deep Reinforcement Learning (DRL) has been used to solve complex robot control tasks with outstanding success. However, previous DRL methods still exist some shortcomings, such as poor generalization performance, which makes policy performance quite sensitive to small vari-ations of the task settings. Besides, it is quite time-consuming and computationally expensive to retrain a new policy from scratch for new tasks, hence restricts the applications of DRL-based methods in the real world. In this work, we propose a novel DRL generalization method called GNN-embedding, which incorporates the robot hardware and the environment simultaneously with GNN-based policy network and learnable embedding vectors of tasks. Thus, it can learn a unified policy for different robots under different environment conditions, which improves the generalization performance of existing DRL robot policies. Multiple experiments on the Hopper-v2 robot are conducted. The experimental results demonstrate the effectiveness and efficiency of GNN-embedding on generalization, including multi-task learning and transfer learning problems.
{"title":"Structure-Aware Policy to Improve Generalization among Various Robots and Environments","authors":"Wei-qing Xu, Yue Gao, Buqing Nie","doi":"10.1109/ROBIO55434.2022.10011945","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10011945","url":null,"abstract":"Recently, Deep Reinforcement Learning (DRL) has been used to solve complex robot control tasks with outstanding success. However, previous DRL methods still exist some shortcomings, such as poor generalization performance, which makes policy performance quite sensitive to small vari-ations of the task settings. Besides, it is quite time-consuming and computationally expensive to retrain a new policy from scratch for new tasks, hence restricts the applications of DRL-based methods in the real world. In this work, we propose a novel DRL generalization method called GNN-embedding, which incorporates the robot hardware and the environment simultaneously with GNN-based policy network and learnable embedding vectors of tasks. Thus, it can learn a unified policy for different robots under different environment conditions, which improves the generalization performance of existing DRL robot policies. Multiple experiments on the Hopper-v2 robot are conducted. The experimental results demonstrate the effectiveness and efficiency of GNN-embedding on generalization, including multi-task learning and transfer learning problems.","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122630566","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-05DOI: 10.1109/ROBIO55434.2022.10011788
Huichen Ma, Junjie Zhou, Lijun Meng, Jianghao Jiang, Sanxi Ma
This paper presents a novel pneumatic soft crawling actuator that exploits scale-like chassis to move. Based on the lateral undulation movement, bellows-type actuators are designed with embedded fluidic chambers that produce bidirectional bending when pressurized. Three chassis structures are created and manufactured to simulate the anisotropy friction by analyzing the legless squamate reptile motion principle. Inspired by the rigid snake robot modeling, a framework to solve the dynamic behavior problem of a soft crawling actuator is further modeled. Particularly, the expected movement has been achieved. Through quantitative analysis, the horizontal belt type shows a more effective drive. Locomotion experimental results of the soft crawling actuator prototype on a carpeted surface show good agreement with model predictions. The demonstrations of terrain adaptability prove movement ability in complicated and constrained environments such as a steep slope, ladders surface, and step surface.
{"title":"Legless Squamate Reptiles Inspired Design: Simple Soft Crawling Actuator","authors":"Huichen Ma, Junjie Zhou, Lijun Meng, Jianghao Jiang, Sanxi Ma","doi":"10.1109/ROBIO55434.2022.10011788","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10011788","url":null,"abstract":"This paper presents a novel pneumatic soft crawling actuator that exploits scale-like chassis to move. Based on the lateral undulation movement, bellows-type actuators are designed with embedded fluidic chambers that produce bidirectional bending when pressurized. Three chassis structures are created and manufactured to simulate the anisotropy friction by analyzing the legless squamate reptile motion principle. Inspired by the rigid snake robot modeling, a framework to solve the dynamic behavior problem of a soft crawling actuator is further modeled. Particularly, the expected movement has been achieved. Through quantitative analysis, the horizontal belt type shows a more effective drive. Locomotion experimental results of the soft crawling actuator prototype on a carpeted surface show good agreement with model predictions. The demonstrations of terrain adaptability prove movement ability in complicated and constrained environments such as a steep slope, ladders surface, and step surface.","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122800597","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Deep learning-based object detection algorithms are gradually promoted in industrial visual detection due to their versatility and high accuracy. These algorithms usually require large amounts of training data, however there is a problem of lack of training samples in actual weld seam detection tasks that challenges the weld seam visual detection task. To improve the performance on weld seam detection, especially for those few-shot tasks, this paper proposes a meta-metric learning method for few-shot weld seam detection. The method introduces a distance metric-learning module besides the meta-learning algorithm. By optimizing the training strategy and classification mode of the base detection model, the method accelerates the training process and improves the learning capability on few-shot weld seam samples. Compared with the base model, the mAP of the method proposed in this paper on the weld seam dataset is improved by about 8.9%.
{"title":"A Deep Meta-Metric Learning Method for Few-Shot Weld Seam Visual Detection","authors":"Tianchen Zhu, Shiqiang Zhu, Jiakai Zhu, Wei Song, Cunjun Li, Hongjiang Ge, Jianjun Gu","doi":"10.1109/ROBIO55434.2022.10012017","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10012017","url":null,"abstract":"Deep learning-based object detection algorithms are gradually promoted in industrial visual detection due to their versatility and high accuracy. These algorithms usually require large amounts of training data, however there is a problem of lack of training samples in actual weld seam detection tasks that challenges the weld seam visual detection task. To improve the performance on weld seam detection, especially for those few-shot tasks, this paper proposes a meta-metric learning method for few-shot weld seam detection. The method introduces a distance metric-learning module besides the meta-learning algorithm. By optimizing the training strategy and classification mode of the base detection model, the method accelerates the training process and improves the learning capability on few-shot weld seam samples. Compared with the base model, the mAP of the method proposed in this paper on the weld seam dataset is improved by about 8.9%.","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121275146","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-05DOI: 10.1109/ROBIO55434.2022.10011688
Yue Gao, Huajian Wu, Mingdong Sun
Legged robot is designed for more flexibility when navigating in complex unstructured environment. When the end-effectors of the robot contacting non-rigid ground, the robot sinks due to different stiffness of the ground. This presents a challenge for accurate and robust control of the upper platform. In this paper, a real-time muti-sensor fusion method Dual Parallelizable Particle Filter (DPPF) is proposed to estimate ground stiffness. DPPF utilized RGB-D camera, IMU and 3-DoF force sensors. Meanwhile, we established a ground material database and trained a real-time ground segmentation network to assist the stiffness estimation of the ground. Then the information of ground material is utilized as a prior distribution for DPPF to achieve faster stiffness estimation. The experiments on synthetic data and on six-legged robot show that DPPF has faster computing speed, fewer convergent steps than previous state estimation methods. The estimated stiffness can be utilized for legged robot impedance control, posture control and trajectory planning.
{"title":"Multi-sensor Fusion for Stiffness Estimation to Assist Legged Robot Control in Unstructured Environment","authors":"Yue Gao, Huajian Wu, Mingdong Sun","doi":"10.1109/ROBIO55434.2022.10011688","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10011688","url":null,"abstract":"Legged robot is designed for more flexibility when navigating in complex unstructured environment. When the end-effectors of the robot contacting non-rigid ground, the robot sinks due to different stiffness of the ground. This presents a challenge for accurate and robust control of the upper platform. In this paper, a real-time muti-sensor fusion method Dual Parallelizable Particle Filter (DPPF) is proposed to estimate ground stiffness. DPPF utilized RGB-D camera, IMU and 3-DoF force sensors. Meanwhile, we established a ground material database and trained a real-time ground segmentation network to assist the stiffness estimation of the ground. Then the information of ground material is utilized as a prior distribution for DPPF to achieve faster stiffness estimation. The experiments on synthetic data and on six-legged robot show that DPPF has faster computing speed, fewer convergent steps than previous state estimation methods. The estimated stiffness can be utilized for legged robot impedance control, posture control and trajectory planning.","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"696 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132795592","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-05DOI: 10.1109/ROBIO55434.2022.10011668
Chengzhi Gao, Ye Xie, Shiqiang Zhu, Guanyu Huang, Lingyu Kong, Anhuan Xie, J. Gu, Dan Zhang, Jun Shao, Haofu Qian
The precise estimation of system states is essential for the locomotion control of biped robots to maintain balance. Currently, the estimation of system states is either based on vision data that is susceptible to the environment, or fusing data from Inertial Measurement Unit (IMU) and the kinematic calculation. Invariant Extended Kalman Filtering (IEKF) is one of the most successful fusing algorithms to estimate system state. Generally, in IEKF, the noise covariance of system state is assumed to be known. However, the noise covariance of contact point for biped is not available since the ground-contact situation normally varies and not previously known. This paper presents a new fusing algorithm-Adaptive Robust Invariant Extended Kalman Filtering (ARIEKF) to adaptively adjust the noise parameter of contact point. The proposed algorithm applied the principle of robust estimation to resist outlier effects of state, and introduced an adaptive factor for the noise covariance of state to control its outlying disturbance influences. This paper firstly completed the full state estimation of biped robot using the theory of Lie groups and invariant observer. Then, the adaptive scale factor evaluated by three-segment approach was adopted to adjust the noise covariance of contact point. Finally, both IEKF and proposed ARIEKF are applied to our biped robot-Cosmos and the accuracy of two algorithms are compared. The mean square errors of the velocity of two algorithms were evaluated using the measurements from motion capture system. Experiments demonstrated that the mean square errors of the velocity are decreased 50 percent when compared with IEKF.
{"title":"Adaptive Robust Invariant Extended Kalman filtering for Biped Robot*","authors":"Chengzhi Gao, Ye Xie, Shiqiang Zhu, Guanyu Huang, Lingyu Kong, Anhuan Xie, J. Gu, Dan Zhang, Jun Shao, Haofu Qian","doi":"10.1109/ROBIO55434.2022.10011668","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10011668","url":null,"abstract":"The precise estimation of system states is essential for the locomotion control of biped robots to maintain balance. Currently, the estimation of system states is either based on vision data that is susceptible to the environment, or fusing data from Inertial Measurement Unit (IMU) and the kinematic calculation. Invariant Extended Kalman Filtering (IEKF) is one of the most successful fusing algorithms to estimate system state. Generally, in IEKF, the noise covariance of system state is assumed to be known. However, the noise covariance of contact point for biped is not available since the ground-contact situation normally varies and not previously known. This paper presents a new fusing algorithm-Adaptive Robust Invariant Extended Kalman Filtering (ARIEKF) to adaptively adjust the noise parameter of contact point. The proposed algorithm applied the principle of robust estimation to resist outlier effects of state, and introduced an adaptive factor for the noise covariance of state to control its outlying disturbance influences. This paper firstly completed the full state estimation of biped robot using the theory of Lie groups and invariant observer. Then, the adaptive scale factor evaluated by three-segment approach was adopted to adjust the noise covariance of contact point. Finally, both IEKF and proposed ARIEKF are applied to our biped robot-Cosmos and the accuracy of two algorithms are compared. The mean square errors of the velocity of two algorithms were evaluated using the measurements from motion capture system. Experiments demonstrated that the mean square errors of the velocity are decreased 50 percent when compared with IEKF.","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"53 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133973682","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-05DOI: 10.1109/ROBIO55434.2022.10011932
Dongni. Yang, Wei Wei, Jiaqian Li, Nan Xiao
Vascular interventional surgery is the most commonly used method for the treatment of cardio-vascular and cerebrovascular diseases. Master-slave interventional surgical robot is a promising technology, which can further improve the accuracy and safety of surgery. However, imperfect measurement of catheter force remains a surgical risk. Inspired by the function of insect antennae, a thin-film force sensing device was installed in the catheter head. Combined with the pressure sensor in the catheter clamping device, the LSTM network was used to predict and classify the curvature of the current passing vessel, and the recognition accuracy was 97%. In the process of robotic surgery, real-time feedback of current pressure information and vascular curvature information can enhance the doctor's judgment of the operation state and improve the safety of surgery.
{"title":"Vascular Environment Identification Based on Multi-dimensional Data Fusion for Interventional Surgical Robots","authors":"Dongni. Yang, Wei Wei, Jiaqian Li, Nan Xiao","doi":"10.1109/ROBIO55434.2022.10011932","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10011932","url":null,"abstract":"Vascular interventional surgery is the most commonly used method for the treatment of cardio-vascular and cerebrovascular diseases. Master-slave interventional surgical robot is a promising technology, which can further improve the accuracy and safety of surgery. However, imperfect measurement of catheter force remains a surgical risk. Inspired by the function of insect antennae, a thin-film force sensing device was installed in the catheter head. Combined with the pressure sensor in the catheter clamping device, the LSTM network was used to predict and classify the curvature of the current passing vessel, and the recognition accuracy was 97%. In the process of robotic surgery, real-time feedback of current pressure information and vascular curvature information can enhance the doctor's judgment of the operation state and improve the safety of surgery.","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"94 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131838099","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-05DOI: 10.1109/ROBIO55434.2022.10012027
Junhang Wei, Shaowei Cui, Peng Hao, Shuo Wang
Endowing robots with tool manipulation skills helps them accomplish challenging tasks. While robots manipulate tools to achieve goals, the alignment of tools and targets is a noise-sensitive and contact-rich task. However, it is difficult to access the accurate pose of the tool and the target. When there is unknown noise in the observations, reinforcement learning can't be sure to perform well. In this paper, we define the easier-to-obtain accurate task-related information as anchor information and introduce a tool manipulation method based on reinforcement learning and anchor information, which can perform well when the observations include unknown noise. To evaluate the method, we build a simulated environment ToolGym, which includes four different kinds of tools and different noise sampling functions for each tool. Finally, we compare our method with baseline methods to show the effectiveness of the proposed method.
{"title":"Learning to Manipulate Tools Using Deep Reinforcement Learning and Anchor Information","authors":"Junhang Wei, Shaowei Cui, Peng Hao, Shuo Wang","doi":"10.1109/ROBIO55434.2022.10012027","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10012027","url":null,"abstract":"Endowing robots with tool manipulation skills helps them accomplish challenging tasks. While robots manipulate tools to achieve goals, the alignment of tools and targets is a noise-sensitive and contact-rich task. However, it is difficult to access the accurate pose of the tool and the target. When there is unknown noise in the observations, reinforcement learning can't be sure to perform well. In this paper, we define the easier-to-obtain accurate task-related information as anchor information and introduce a tool manipulation method based on reinforcement learning and anchor information, which can perform well when the observations include unknown noise. To evaluate the method, we build a simulated environment ToolGym, which includes four different kinds of tools and different noise sampling functions for each tool. Finally, we compare our method with baseline methods to show the effectiveness of the proposed method.","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134189177","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-05DOI: 10.1109/ROBIO55434.2022.10011837
Senwei Xiang, Minxiang Ye, Shiqiang Zhu, J. Gu, Anhuan Xie, Zehua Men
Electric vertical takeoff and landing aircraft (eVTOL) has drawn more and more attention from home and abroad in recent years. It is believed autonomous eVTOL will create a new era of Urban Air Mobility (UAM) and Advanced Air Mobility (AAM). To autonomous eVTOL, precision landing is an extremely critical operation for it directly affects flight safety. In this paper, we analyze the special issues an eVTOL will encounter when it lands in the UAM applications. A multi-stage precision landing method based on multi-marker joint localization is proposed accordingly. Our method contains three key elements: a compatible vertiport with multiple visual markers; an accurate, fast detection and localization algorithm for the vertiport and a multi-stage landing strategy. We implement our method on an eVTOL prototype named ZJ-Copter developed by Zhejiang Lab. A series of real-world experiments have been conducted to validate the effectiveness and accuracy of the proposed method. Experiment results show that our method works well in real-world scenarios for autonomous eVTOL without GPS signal during landing process. The positioning accuracy is less than 0.1m (altitude $< mathbf{10}mathbf{m}$), while the landing accuracy is less than 0.5m.
{"title":"A Multi-stage Precision Landing Method for Autonomous eVTOL Based on Multi-marker Joint Localization","authors":"Senwei Xiang, Minxiang Ye, Shiqiang Zhu, J. Gu, Anhuan Xie, Zehua Men","doi":"10.1109/ROBIO55434.2022.10011837","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10011837","url":null,"abstract":"Electric vertical takeoff and landing aircraft (eVTOL) has drawn more and more attention from home and abroad in recent years. It is believed autonomous eVTOL will create a new era of Urban Air Mobility (UAM) and Advanced Air Mobility (AAM). To autonomous eVTOL, precision landing is an extremely critical operation for it directly affects flight safety. In this paper, we analyze the special issues an eVTOL will encounter when it lands in the UAM applications. A multi-stage precision landing method based on multi-marker joint localization is proposed accordingly. Our method contains three key elements: a compatible vertiport with multiple visual markers; an accurate, fast detection and localization algorithm for the vertiport and a multi-stage landing strategy. We implement our method on an eVTOL prototype named ZJ-Copter developed by Zhejiang Lab. A series of real-world experiments have been conducted to validate the effectiveness and accuracy of the proposed method. Experiment results show that our method works well in real-world scenarios for autonomous eVTOL without GPS signal during landing process. The positioning accuracy is less than 0.1m (altitude $< mathbf{10}mathbf{m}$), while the landing accuracy is less than 0.5m.","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130308015","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-05DOI: 10.1109/ROBIO55434.2022.10011748
Li Wang, Wenbo Shi, Haozhe Zhu, D. Zhang, Yikang Zhang, Jiahe Fan, M. J. Bocus
Road defects can severely affect the safety of road users and vehicle conditions. Over the past decade, due to the limited amount of labeled training data, machine vision-based road defect detection approaches have been mainly used, while machine/deep learning-based methods were merely discussed. With the recent development of artificial intelligence, convolutional neural network (CNN)-based road defect detection systems for automated road condition assessment have become an active sphere of study. In this regard, this paper presents a comprehensive road defect detection system based on computer stereo vision, non-linear regression, and CNN. A dense disparity image is first estimated from a pair of stereo road images using an efficient stereo matching algorithm. The estimated disparity image is then transformed to better identify road defects by minimizing a global energy function w.r.t. road disparity projection model coefficients and stereo rig roll angle, using the non-linear regression approach. Finally, three popular semantic segmentation CNNs are trained using the transformed disparity images. Extensive experiments are conducted to demonstrate the performance of our proposed road defect detection approach. The achieved pixel-level accuracy and intersection over union (IoU) are 98.37% and 67.65%, respectively.
{"title":"Road Defect Detection Based on Semantic Transformed Disparity Image Segmentation","authors":"Li Wang, Wenbo Shi, Haozhe Zhu, D. Zhang, Yikang Zhang, Jiahe Fan, M. J. Bocus","doi":"10.1109/ROBIO55434.2022.10011748","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10011748","url":null,"abstract":"Road defects can severely affect the safety of road users and vehicle conditions. Over the past decade, due to the limited amount of labeled training data, machine vision-based road defect detection approaches have been mainly used, while machine/deep learning-based methods were merely discussed. With the recent development of artificial intelligence, convolutional neural network (CNN)-based road defect detection systems for automated road condition assessment have become an active sphere of study. In this regard, this paper presents a comprehensive road defect detection system based on computer stereo vision, non-linear regression, and CNN. A dense disparity image is first estimated from a pair of stereo road images using an efficient stereo matching algorithm. The estimated disparity image is then transformed to better identify road defects by minimizing a global energy function w.r.t. road disparity projection model coefficients and stereo rig roll angle, using the non-linear regression approach. Finally, three popular semantic segmentation CNNs are trained using the transformed disparity images. Extensive experiments are conducted to demonstrate the performance of our proposed road defect detection approach. The achieved pixel-level accuracy and intersection over union (IoU) are 98.37% and 67.65%, respectively.","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130354545","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The traditional dual manipulator control systems have not only complex motion coupling problems, but also larger computational burden, and hence it is difficult to meet the requirements of intelligent assembly. In this paper, based on multi-agent reinforcement learning theory, Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm is investigated in the collaborative assembly shaft slot assembly via dual manipulator system. For the collaborative shaft slot assembly in the dual manipulator system, sparse rewards in traditional multi-agent reinforcement learning often exist because of the long sequence decision-making problem. For the above problems, this paper considers the influence of the decision-making of a single manipulator on the overall task rewards when the overall rewards of multi -agent reinforcement learning are designed. In the proposed algorithm, by calculating the difference before and after the state of each manipulator, and applying the difference as the internal state excitation to the overall task rewards, the traditional reward function of multi-agent reinforcement learning is improved. In order to verify the designed algorithm, the dual manipulator shaft slot assembly system and test scenario are established on the CoppeliaSim simulation platform. Simulation results show that the success rate of the shaft slot assembly via the improved MADDPG algorithm is about 83 % *
{"title":"Dual manipulator collaborative shaft slot assembly via MADDPG","authors":"Junying Yao, Xiaojuan Wang, Renqiang Li, Wenxiao Wang, X. Ping, Yongkui Liu","doi":"10.1109/ROBIO55434.2022.10011768","DOIUrl":"https://doi.org/10.1109/ROBIO55434.2022.10011768","url":null,"abstract":"The traditional dual manipulator control systems have not only complex motion coupling problems, but also larger computational burden, and hence it is difficult to meet the requirements of intelligent assembly. In this paper, based on multi-agent reinforcement learning theory, Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm is investigated in the collaborative assembly shaft slot assembly via dual manipulator system. For the collaborative shaft slot assembly in the dual manipulator system, sparse rewards in traditional multi-agent reinforcement learning often exist because of the long sequence decision-making problem. For the above problems, this paper considers the influence of the decision-making of a single manipulator on the overall task rewards when the overall rewards of multi -agent reinforcement learning are designed. In the proposed algorithm, by calculating the difference before and after the state of each manipulator, and applying the difference as the internal state excitation to the overall task rewards, the traditional reward function of multi-agent reinforcement learning is improved. In order to verify the designed algorithm, the dual manipulator shaft slot assembly system and test scenario are established on the CoppeliaSim simulation platform. Simulation results show that the success rate of the shaft slot assembly via the improved MADDPG algorithm is about 83 % *","PeriodicalId":151112,"journal":{"name":"2022 IEEE International Conference on Robotics and Biomimetics (ROBIO)","volume":"318 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115026022","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：