Study On Grasp Force Control Of Agricultural Robot End-Effector

The success rate of fruit and vegetable picking with the agricultural robot is very low and the loss rate is very high, because of task complexity and environment particularity. A major reason of high damage rate is that grasp force on end-effector is too big or small, which makes the fruit and vegetable crack or fall off. Grasp force can turn small gradually while grasping fruit and vegetable, it is necessary to consider the deformation effect on grasp force. It was required that the end-effectors were provided with compliance for no any damage in the course of grasping fruit and vegetable. Active Compliance control, that is, force control was adopted in order to solve the damage while manipulator picking or grasping fruits and vegetables. The completed work is generalized as follows:1. A five-component mechanical model was built with the springs, dampers, and torsional springs through series parallel of them, based on agricultural rheology, robotics, robotic manipulation. Given that the distribution distance between robot fingers and object is significantly larger than dimensions of fingers, each finger was considered to be contacted with a five-component model, thus a general static grasp model of fruit and vegetable was established by rigid grasping theory. Grasp map matrix was derived for robot grasping fruit, and the required grasp condition was calculated according to bounded force closure.2. An experiment platform of automatic grasping system for fruit and vegetable was developed. The experiment platform was consisted of end-effector, control card, data acquisition card, AC servo motor, AC servo driver, force sensor, signal amplifier, regulated power, industrial computer, et al, so the open architecture force control system of end-effector was built up based on PC. The grasp system has good expansibility, generality, practicability, and force control precision about end-effector can be satisfied.3. Deformation can occur while end-effector grasping fruit and vegetable, so series control with force/position control was adopted in the control system frame. The external force control loops algorithm was proposed for the control system in experiment, and force error was converted to speed input of position control system. The end-effector speed adaptive decelerated as force error decreasing in order to reduce force overshoots. The grasp control system is steady, the measuring precision of force sensors is0.3%FS, the success rate of grasping tomatoes is beyond96%, and the damage rate is less than10%, and all the damages appear in high speed grasping.4. Incremental PI force control algorithm was proposed based on grey prediction in order to make grasp force track set value quickly without big overshoot. Grey prediction model was built by force data acquisition from the grasped objects, the weights of predictive force error were increased or decreased in integrated error accordingly to that the precision of predictive model is high or low. Force controller can employ the past, present and future grasp force information to calculate an appropriate control correction to pre-compensate the force error, and can yield small overshot, fast response simultaneously, which make the controller adaptive to the dynamic grasp process between deformable objects and end-effector. Experimental results are presented to demonstrate the efficacy of grey predictive incremental PI force control algorithm, which can reduce the damage rate in grasping.5. If experience and knowledge about human grasping fruit and vegetable can be translated into control rules, successful and safe grasping of end-effectors would be increased evidently. Conventional control can’t solve force control satisfactorily because stiffness of fruits and vegetables is different. The force controller designed with standard condition may be instability when stiffness of fruits and vegetables is quite high or low. Composite controller consisted of fuzzy controller and integral controller was used to improve control performance. Integral controller could eliminate or reduce steady-state error, and improved the steady performance. A concurrent control force algorithm was proposed and tested as a good method to grasp fruit and vegetable.6. Prediction control algorithm is suitable for manipulator grasping with rotary joint, too. The relation between joint torque and joint angle was built through angular stiffness, the change of joint torque was predicted to help to know grasping force information to decrease the loss when grasping fruit and vegetable. Grasping control schemes was proposed based on external torque control loops, and the future joint torque would be gradually added to torque loops according to sampling period based on linear interpolation between the current torque and predictive one. Force overshoot and oscillation about output torque and angular velocity of manipulator would be depressed, made the grasped objects free from damage.7. Agriculture harvesting robot always works under high non-normal environment. In order to suppress the effect of process and measurement to control system, a RBF neural network and PD combined controller was proposed based on the Kalman filter, the controller was composed of RBF neural network forward feed controller, PD feedback controller and Kalman filter. It always could fast track the fixed value when the noise amplitude was enlarged or sampling frequency of control system was changed, which has excellent tracking performance and quite adaptive and robust.