Cai Hegao

Study on a SVM-based data fusion method

A new two-stage SVM-based data fusion strategy is proposed and it is applied to obtain the accurate information of the robot gripper state. Support vector machines (SVM) operate on the principle of structure risk minimization which not only keeps the empirical risk minimal but also control VC confidence of discriminate functions, hence better generalization ability is guaranteed. In this paper, the basic principles of SVM are discussed first and then a classified and graded data fusion strategy is proposed according to the features of the problem of gripper information data fusion. Finally, experimental results demonstrate the advantages and efficiency of the proposed approach.

Real-time self-reaction of mobile robot with genetic fuzzy neural network in unknown environment

Presents an intelligent control method for real-time self-reaction of a mobile robot in an unknown environment, it is called a genetic fuzzy neural network. It is used to control a mobile robot according to sensing different information, which includes the different direction distances between the obstacles and robot sensed by ultrasonic sensors, the target orientation sensed by a simple optical range-finder and the robots direction of movement. In the paper, the distances d/sub r/, d/sub c/, and d/sub l/ between the robot and the obstacles with respect to the right, front and left sensors, as well as the angle t/sub r/, between the target orientation and the robots direction of movement are taken as the inputs of the intelligent controller. The output of the intelligent controller is the steering angle s/sub a/. A genetic fuzzy neural network is presented to describe the fuzzy reasoning relationship between the inputs and the output of the system. It has a few advantages, such as higher learning speed and easier ensuring convergence. Simulation results of mobile robot collision avoidance in an unknown environment show the method presented is feasible.

The path planning of mobile manipulator with genetic-fuzzy controller in flexible manufacturing cell

This paper presents a new intelligent path planning method for motions of the mobile manipulator in flexible manufacturing cell. It simulates the way of human actions to control the mobile manipulator according to a master-slave hierarchical control strategy. The method can find path using the genetic fuzzy controller (GFC) while in moving, and switch to the collision-free path planner based on the critical collision joint angle (CCJA) and DA* algorithms when in requirement of performing the tasks such as load or unload. In this paper, the GFC is developed to automatically generate the fuzzy if-then rules from samples. In addition, a new modeling approach for manipulator is presented. It is called the c-space modeling based on CCJA. Using this technique, the computation time and memory space are greatly reduced. Moreover, a modified DA* algorithm with dynamic step-changeable and goal-visible-test is also used to accelerate the search process. Simulation results of the mobile manipulator performing object transporting task show the method presented is feasible.

Capture Strategy Based on a Distributed Control System for a Large-scale Space End-Effector

For the positioning problem of a large-scale space end-effector from the flexible joints and elastic beams before the capture operation,a distributed control system is proposed for a newly developed end-effector equipped with flexible capture interface.A capture-snare-speed-oriented strategy is established based on distributed control system;thus,a reliable and effective capture operation can be realized.The designed system is composed of a sensor unit,a top-level control unit,and three modularized bottom-level driving units.The sensor unit and the bottom-level driving unit are integrated inside the end-effector, and used to obtain the data of position,force/moment,current and temperature sensors and drive the servo motors of the system.The top-level control unit sends control commands and parameters synchronically to the three bottom-level driving units through a high-speed bus for coordinating the catching,dragging and latching operations of the end-effector. Experimental results show that,in support of the distributed control system,the reliability of the capture operation can be effectively improved by using the proposed strategy.

Autonomous Kinematic Self-Calibration of a Novel Haptic Device

A closed loop self-calibration method for Delta mechanism and 3-RRR mechanism used in a double parallel haptic device was presented. This approach blocks successively one joint of each parallel mechanism using a simple lockup device and reforms the 3-DOF mechanism to be a 2-DOF one. The parameter errors of the mechanism are identified according to the differences between the sensor output of the redundant chain and nominal model output. Both the condition number and singular value of the error propagation Jacobian are synthesized as a new selection principle of calibration set such that the calibrating process is insensitive to the noise and the calibration precision is increased efficiently. Experiments show that the ratio of maximal error before and after calibrating is 9.6, namely, the setting accuracy of the two mechanisms are increased about 10 times and the error of re-position is less than 0.1 mm and 0.1deg

Safety in multi-robot teleoperation system through Internet

Internet based multi-operator multi-robot (MOMR) teleoperation system allow the accomplishment of tasks that a single robot could not approach due to restrictions associated to the product and /or the robot. The safety issue in the system is an important problem according to its complexity. This paper describes the development of multi-layer safety structure of the safety subsystem which consists of software modules as well as hardware components after the threat analysis of the system using fault tree analysis (FTA) technology. The subsystem is divided into three layers with its logic architecture: interactive monitor layer, collaborative control layer and real-time control layer. The safety problems and the related strategy are clarified by detailed analysis of each layer and relationship among the layers. So we can get a high performance MOMR teleoperation system with multi-layer safety architecture

Foot trajectory planning of frog swimming based on propulsion mechanism

To provide reasonable reference trajectories for frog inspired robot, the bionic trajectory generation was studied in this paper. Though joint trajectories vary with different motion modes, joint speeds are too fast and tuned with the musculoskeletal system. However, mechanical structures of bionic robot are simplified and diver abilities limit the speed, so the method to generate joint trajectories which not only reflect the characteristics of real frog swimming but also are within the limitation of drivers were proposed in this paper. Firstly, swimming mechanism of frog was briefly introduced to analyze motion patterns of hind limbs and webbed feet. Secondly, trajectory (relative to body) planning was divided into hind limb planning and webbed foot planning. Hind limb trajectory planning was conducted via end track planning and acceleration planning, while acceleration planning was operated for webbed foot planning. Then, trajectories could be operated from fast to slow by adjusting duration time of the planned trajectories. Finally, the trajectories were used as input of hydrodynamic simulation to validate the trajectory planning method.

Genetic algorithm based path planning of coordinated multi-robot manipulators

The path planning of coordinated multi-robot is a complicated problem, which involves the valid search in a high-dimension space. In terms of concrete operation situations, two types of operation of loose coordination and tight coordination of multi-robot manipulators are analyzed in detail, respectively. Furthermore, the varying length string and the fitness evaluation of GA are also discussed. Finally, two types of representative coordination cases are given to validate the proposed approach.

The implementation of active/passive coordinated manipulation in precision assembly based on force-feedback

The authors present a new idea on assembly principle and a newly developed robotic assembly system based on force feedback, which employs a specially designed active/passive micro manipulator (APMM) to combine both active accommodation and passive compliance function in one. The test setup consists of a PUMA 562 robot, an APMM and its driver unit, a 6 dimensional force/torque sensor and its controller, an IBM-PC computer and specially designed interfaces. During assembly, the PUMA robot takes care of coarse positioning and insert motion, while the APMM, which is mounted on the end of PUMA robot, executes fine position adjustment by optimal cooperation of active and passive mating based upon force feedback. The system can carry out assembly operation along arbitrary direction within several seconds even under the condition of unchamfered parts.<<ETX>>

Identification and smooth tracking control of robot force control system

In order to obtain high performance of the force control system during deburring operation, the new adaptive feedforward force control scheme (adaptive zero phase error tracking control) to achieve accurate tracking contact force which is based on the ARMAX plant models identified using the least squares method is proposed. Simulation results of a robotic deburring system are given to illustrate the efficiency of the proposed scheme.<<ETX>>