The current research of quadruped robot focuses on the quadruped robot with spine motion. Contact time is a very important part of system performance. However, the mechanism of spine motion about contact time has not ...The current research of quadruped robot focuses on the quadruped robot with spine motion. Contact time is a very important part of system performance. However, the mechanism of spine motion about contact time has not been clearly elucidated. In this paper, the e ect of spine motion on contact time is studied deeply from dynamic view.Firstly, a simplified model of the quadruped robot with spine joint is set up, its dynamic equations are derivated, and a method that can generate passive periodic locomotion is proposed. Secondly, according to the vertical spring oscillator model, the two-dimension planar locomotion of the simplified model is regarded as a special vibration in the vertical direction, and the approximate formula of calculating contact time is obtained. Finally, the approximate formula of calculating contact time is verified by the simulation results of passive periodic locomotion, and the e ect of spine motion on contact time is deeply discussed based on the approximate formula of calculating contact time. The discussion proves that spine motion indeed has little e ect on contact time, but spine motion can slightly reduce body pith movement and regulate the leg sti ness in leg contact phase. This research proposes an e ective research method which can be used to study the motion mechanism of the quadruped robot with spine motion,and the mechanism of spine motion about contact time is clearly elucidated which is helpful to set the parameters of mechanical structure and study control algorithm about the quadruped robot with spine motion.展开更多
为解决煤矿巷道环境恶劣及人工疲劳驾驶电机车导致煤矿井下有轨电机车事故频发的问题,提出一种基于改进YOLOv4-Tiny算法的YOLOv4-Tiny-4S矿井电机车多目标实时检测方法。首先,为了提高网络模型对于小目标的检测能力,将传统YOLOv4-Tiny...为解决煤矿巷道环境恶劣及人工疲劳驾驶电机车导致煤矿井下有轨电机车事故频发的问题,提出一种基于改进YOLOv4-Tiny算法的YOLOv4-Tiny-4S矿井电机车多目标实时检测方法。首先,为了提高网络模型对于小目标的检测能力,将传统YOLOv4-Tiny的两尺度预测增加至4尺度预测,并且在网络模型的颈部引入空间金字塔池化(spatial pyramid pooling,SPP)模块,以丰富特征融合信息,增大网络模型的感受野。其次,以煤矿巷道中的行人、电机车、信号灯及碎石作为检测目标,创建矿井电机车多目标检测数据集,并分别采用K-means和K-means++聚类分析算法对数据集重新聚类;对比分析结果表明,K-means++算法具有更好的聚类效果。最后,通过对传统YOLOv4-Tiny算法的消融实验,进一步展示了不同改进措施对网络模型检测性能的影响;并在电机车运行的煤矿巷道场景中,对比分析了YOLOv4-Tiny-4S算法与其他几种算法的检测性能。实验结果表明:YOLOv4-Tiny-4S算法能够准确检测并识别出图像中的各类目标,其平均精度均值(mean average precision,mAP)为95.35%,对小目标“碎石”的平均精度(average precision,AP)为86.69%,相比传统YOLOv4-Tiny算法分别提高了12.38%和41.66%;改进后算法的平均检测速度达58.7帧/s(frames per second,FPS),模型内存仅为26.3 Mb,YOLOv4-Tiny-4S算法的检测性能优于其他算法。本文提出的基于YOLOv4-Tiny-4S矿井电机车多目标实时检测方法可为实现矿井电机车的无人驾驶提供技术支撑。展开更多
基金Supported by National Natural Science Foundation of China and China Academy of Engineering Physics(NSAF,Grant No.U1530119)
文摘The current research of quadruped robot focuses on the quadruped robot with spine motion. Contact time is a very important part of system performance. However, the mechanism of spine motion about contact time has not been clearly elucidated. In this paper, the e ect of spine motion on contact time is studied deeply from dynamic view.Firstly, a simplified model of the quadruped robot with spine joint is set up, its dynamic equations are derivated, and a method that can generate passive periodic locomotion is proposed. Secondly, according to the vertical spring oscillator model, the two-dimension planar locomotion of the simplified model is regarded as a special vibration in the vertical direction, and the approximate formula of calculating contact time is obtained. Finally, the approximate formula of calculating contact time is verified by the simulation results of passive periodic locomotion, and the e ect of spine motion on contact time is deeply discussed based on the approximate formula of calculating contact time. The discussion proves that spine motion indeed has little e ect on contact time, but spine motion can slightly reduce body pith movement and regulate the leg sti ness in leg contact phase. This research proposes an e ective research method which can be used to study the motion mechanism of the quadruped robot with spine motion,and the mechanism of spine motion about contact time is clearly elucidated which is helpful to set the parameters of mechanical structure and study control algorithm about the quadruped robot with spine motion.
文摘为解决煤矿巷道环境恶劣及人工疲劳驾驶电机车导致煤矿井下有轨电机车事故频发的问题,提出一种基于改进YOLOv4-Tiny算法的YOLOv4-Tiny-4S矿井电机车多目标实时检测方法。首先,为了提高网络模型对于小目标的检测能力,将传统YOLOv4-Tiny的两尺度预测增加至4尺度预测,并且在网络模型的颈部引入空间金字塔池化(spatial pyramid pooling,SPP)模块,以丰富特征融合信息,增大网络模型的感受野。其次,以煤矿巷道中的行人、电机车、信号灯及碎石作为检测目标,创建矿井电机车多目标检测数据集,并分别采用K-means和K-means++聚类分析算法对数据集重新聚类;对比分析结果表明,K-means++算法具有更好的聚类效果。最后,通过对传统YOLOv4-Tiny算法的消融实验,进一步展示了不同改进措施对网络模型检测性能的影响;并在电机车运行的煤矿巷道场景中,对比分析了YOLOv4-Tiny-4S算法与其他几种算法的检测性能。实验结果表明:YOLOv4-Tiny-4S算法能够准确检测并识别出图像中的各类目标,其平均精度均值(mean average precision,mAP)为95.35%,对小目标“碎石”的平均精度(average precision,AP)为86.69%,相比传统YOLOv4-Tiny算法分别提高了12.38%和41.66%;改进后算法的平均检测速度达58.7帧/s(frames per second,FPS),模型内存仅为26.3 Mb,YOLOv4-Tiny-4S算法的检测性能优于其他算法。本文提出的基于YOLOv4-Tiny-4S矿井电机车多目标实时检测方法可为实现矿井电机车的无人驾驶提供技术支撑。
