Research On The Uncalibrated Dynamic Visual Servoing Of Robotic Manipulators

Visual servoing of robot manipulators is a key approach that uses visual information from a vision system for feedback control of the motion of robot manipulators. Compared with traditional control approaches, it can provide higher design flexibility, task precision, and intelligence levels. Therefore, it attracts extensive attention, and is an important re-search direction in robotics. Compared with kinematics-based calibrated visual servoing, uncalibrated dynamic visual servoing does not require precise calibration for system param-eters such that the tediousness and difficulty of precise calibration can be avoided, and can guarantee very high system performance even when only approximate values, best estimated values, or even arbitrary estimated values of the system parameters are provided. Further-more, since the nonlinear dynamics of robot manipulators is taken into account, it can also meet the requirements of high-speed and high-performance visual servoing, and hence is a promising visual servoing approach and has become a hot research topic of visual servo-ing in recent years. Based on the existing research results, we conduct further and in-depth investigations on the uncalibrated dynamic visual servoing in this dissertation.Firstly, basic problems of traditional visual servoing are introduced, and research re-sults on uncalibrated visual servoing are systematically summarized. Then, the most funda-mental kinemaitcs modeling methods in visual servoing are discussed, and the kinematics models of both position-based and image-based visual servoing are derived in detail for both types of fixed-camera and eye-in-hand camera vision systems such that we can get a deeper understanding of the key issues of visual servoing and design various types of visual servoing controllers more conveniently. Also, this can provide us with a solid theoretical background for the studies in this dissertation.Existing approximate Jacobian control approaches can still guarantee the asymptotical stability of robot systems when only approximate values or best estimated values of system parameters are used, but they all need to use exact joint velocity measurements. To remove this requirement, we propose an approximate Jacobian control strategy without the use of exact joint velocity measurements, and two sliding-mode-observer strategies are designed to estimate the joint velocities online and their corresponding stability problem is analyzed in detail as well. The proposed approach can improve the control performance of visual regulation systems in real environments.Existing adaptive Jacobian-based task-space tracking control approaches can simulta-neously cope with uncertainties of kinematics and dynamics, but they need to use exact measurements of either task-space velocities, joint-space velocities, or even both of them such that the system performance can be largely affected by measurement noises in real ap-plications. To further improve control performance, we propose an adaptive visual tracking approach without the use of both the task-space and joint-space velocity measurements, and three sliding-mode-observer strategies are designed to estimate the joint velocities online and their corresponding stability problem is analyzed in detail as well.Different from the adaptive Jacobian-based adaptive visual tracking approaches which can only be applied to the cases with time-invariant or slowly time-varying depth parame-ters, the depth-independent Jacobian-based uncalibrated visual servoing can deal with the most general 3-D motion problem of robot manipulators with no constraints on the depth parameters, and can also handle uncertainties of camera intrinsic and extrinsic parameters, and so is a promising visual servoing approach. In the existing depth-independent Jacobian-based fixed-camera visual tracking approach that uses multiple feature points, the designed adaptive law for the estimation of unknown camera parameters needs to use exact values of depth parameters which cannot be obtained, and hence in the implementation stage, the ex-act values should be replaced by their estimated values. In this manner, however, the asymp-totical stability or even stability of robot systems can no longer be theoretically guaranteed. To solve this problem, we propose an improved adaptive law, which is designed by using the estimated values of depth parameters but not their exact values, and on this basis, an initial-state-independent adaptive law is also presented. Both of the proposed adaptive laws can theoretically guarantee the asymptotical stability of robot systems. To handle uncertain-ties of coordinate parameters of the feature points in a fixed-camera visual servoing system, we propose a new depth-independent Jacobian-based kinematics modeling approach, which can cope with the uncertainties of both camera parameters and coordinate parameters of the feature points. On this basis, we propose a unified kinematics-modeling framework, which can be simultaneously applied to both the fixed-camera and eye-in-hand camera vision sys-tems, and then both the visual regulation and visual tracking controller designs are discussed based on this unified framework. To cope with the over-parameterization problem in the uni-fied framework, we design a new depth-independent Jacobian-based parameter estimation strategy, which can be used to decouple the estimation of camera parameters from that of coordinate parameters of feature points, such that the number of parameters to be estimated and the computational complexity can both be reduced.Finally, to solve the consistency problem of desired image trajectories, we propose a new uncalibrated image-space path planning algorithm for the fixed-camera setup, which can generate intermediate reference image features for connecting the initial and desired image features without the knowledge of camera intrinsic parameters and 3D object model. By taking into account the robot dynamics, asymptotical stability of the robot system under the control of the proposed approaches is proved by using Lyapunov theory, and the effectiveness of the proposed approaches is validated by simulation and preliminary experimental studies.