Visual sensor is one of the most essential components for a robot control system. The visual information is expected to be fast, accurate and reliable in providing real-time information of dynamical surroundings. Due to the limitation of computation capability, the performance of vision-based control systems, in particular of highly dynamic visual servo control systems, is often impacted by the low sampling rate of the visual feedback.
In order to overcome the low sampling rate problem, networked visual servo control, which integrates networked computational resources for parallel image processing, is considered for high-speed and high-performance vision-based control due to its merits of easy maintenance, high flexibility and robustness. In contrast to conventional visual servoing, a communication network is involved in the feedback loop for large volume image data exchange, which challenges the control design and the communication strategy. Besides, from the advances of research in neurobiology, biologically inspired vision system attracts increasing attention. The insect-inspired vision is computationally cheap, particularly suitable for real-time applications, and to date, however, less applied for closed loop control.
This thesis provides a comprehensive development concept for a high-speed networked visual servo control system, which merges different design issues including communication, control performance, stability, and network usage. A real-time transport protocol is developed for large volume image data transmission and thus, a cloud computing platform is established enabling high sampling rate visual feedback. A novel switching control law is proposed based on the analysis of sampled-data stochastic systems with time-varying feedback time delays. With regard to limited communication resources the cost-performance trade-off is addressed by innovative strategies of sending rate scheduling. Besides, this thesis summarizes guidelines for the design of an insect-inspired high-speed vision system for robot control. The emphasis lies on the accurate motion estimation and the stability for the closed loop system. The intrinsic characteristics of the insect-inspired motion detector are extensively investigated to design a motion estimation algorithm based on lookup table and a stable system with high feedback gains and delays. The proposed approaches are validated by experiments conducted on seven/one degrees-of-freedom robotic manipulators. The experimental results show superior performance compared to the conventional counterpart, and provide valuable insights for the future work in the framework of high-speed visual servo control.
«
Visual sensor is one of the most essential components for a robot control system. The visual information is expected to be fast, accurate and reliable in providing real-time information of dynamical surroundings. Due to the limitation of computation capability, the performance of vision-based control systems, in particular of highly dynamic visual servo control systems, is often impacted by the low sampling rate of the visual feedback.
In order to overcome the low sampling rate problem, netw...
»