Feature Completeness And Robust Control In Robotic Uncalibrated Visual Servoing

Visual information is introduced into robotic closed control loop in robotic visual servoing.In this way,system states are sensed from the outside,describing the whole task in visual space.When the visual feature set is transformed from initial value into expected value,the task is finished.Uncalibrated visual servoing systems carry out this procedure with uncalibrated cameras.Different feature sets have been proposed for specific tasks to achieve different goals.Few general theories are proposed to evaluate visual features.Completeness is the precondition of improving task performance.Therefore,a general theory is necessary to choose visual features from the angle of visual servoing task completeness,to build hand-eye relationship and to optimize system performance.A visual feature set is required to fully represent each control freedom degree of the robot.On the other hand,for a complicated visual servoing task,describer is also required to characterize the task goal and constraints in high level.Considering this necessity,feature sets are investigated from local and global angles,respectively.Firstly,the concept of complete feature set of visual servoing is proposed.Then,completeness judging methods for a local feature set and a global one are proposed,respectively.In addition,how to obtain and make use of complete feature sets is studied.Last but not least,a robust visual servoing method is introduced to optimally accomplish a visual servoing task with a complete feature set.The main research work and contributions of this thesis are summarized as follows:(1)The concept of feature completeness of visual servoing is proposed.An evaluation technique of a complete local feature set is introduced from the specific control point of view.How to capture and utilize a complete local feature set is investigated in the background of grasping brush task with a humanoid robot.An image Jacobian matrix is utilized to characterize the hand-eye relationship.Accordingly,the rank of an image Jacobian matrix is calculated to determine whether the corresponding feature set represent all the control freedom degrees,i.e.,the completeness of the local feature set.CAMShift algorithm is utilized to robustly capture a complete local feature set,based on which Kalman-Bucy filter based uncalibrated visual servoing method is made used of to accomplish a visual servo task.(2)The concept of complete global feature set is introduced from the overall task goal and constraints point of view.Taking the task of robotic Chinese calligraphy as example,how to obtain and utilize a complete global feature set is studied.A feature set,which can fully characterize the task goal and constraints of high level,is defined as a complete global feature set.Because of the task complexity,only partial features of the complete feature set are directly gotten from images while the other features are reasoned out based information from images.Stroke features based C4.5 decision tree is introduced to obtain stroke direction while heuristic hierarchical reasoning method is proposed to get stroke order.Based on an obtained complete feature set,robotic Chinese calligraphy is fulfilled optimally through aesthetic evaluation based online planning.(3)Based on a complete feature set,which ensures the task realizability,a robust uncalibrated visual servoing method based on disturbance observer is proposed to optimally fulfill a visual servoing task.There exists the problem of singularity,local minima,and external disturbances in traditional uncalibrated visual servoing techniques.To solve this problem,a two-closed-loop structure is built based on disturbance observer.Through designing a proper Q-filter in the inner loop,inner stability of the system is ensured while equivalent disturbances involving model uncertainties,external disturbances and sensing noises are eliminated.In this way,the inner loop equals to a plant described by a given nominal model.so that the outer-loop control is designed directly based on the nominal model to achieve optimal dynamic and robust performance.