Human-Machine Robot Control System Parameter Identification

Abstract

To complete the control of exoskeleton carrying robot perfectly, the human-machine interaction forces model should be identified, which can be simulated using spring-damper model, that is, the coefficient elasticity and damping should be gotten. For the coupling of the several joints, the parameters should be optimized from the system global performance. In this paper, Estimation of Distribution Algorithm(EDA) is used to identification interaction parameters. Second-order EDA based on general structure Gauss network is introduced to replace the condition probability density function, the crossover and mutation operators are added to speed the evolution process. Combining the individual energy-entropy selection, the detail human-machine interaction forces identification method using the improved estimation of distribution algorithm is given and the human-machine carrying robot control system simulation results show the validity of the method. Lower extremity exoskeleton intelligent carrying system is a new concept human-machine intelligent robot system(1). It has two mechanic legs and is attached to the human at various points along the human's and upper body and assists the human to carry the load mounted on its torso. Operator is able to perform a wide range of physical activities while wearing the device. The exoskeleton shadows the motions of the human and never interferes with these motions. For the dynamic repeat characteristics of this system, previously , (2) has shown that iterative learning control is propitious to control the system. The greatest difference between exoskeleton robot and other robot on iterative learning control lies in the existence of the human-machine interaction forces in the carrying robot system. To complete the iterative learning control of exoskeleton robot perfectly, the human- machine interaction forces model should be identified, which can be simulated using spring-damper model, that is, the coefficient elasticity and damping should be gotten. For the coupling of the several joints, the human-machine interaction forces can not be identified purely from a certain joint, so that the parameters should be optimized from the system global performance. The conventional system identification method such as least square method can not get good result, while some new system identification method such as wavelet network method, genetic algorithm and fuzzy logic can overcome the defaults of the conventional method. In the aspect of solving the combination and parameter optimization problem, genetic algorithm is better, which is used to many fields successfully. At the same time, the genetic algorithm also has some defects. For example, it always falls into a local optimal solution and the local search ability is bad. The system information also can't be used fully. In this paper, estimation of distribution algorithm is introduced to to identification interaction parameters. EDA presents a new evolutionary pattern. There is no traditional genetic operation such as crossover and mutation, but the learning and sampling of the probabilistic model. EDA describe the distribution of the candidate solutions in space through a probability model, uses statistical learning tools from the perspective of the macro group to establish a probabilistic model to describe the distribution of solutions, and then take a random sampling of the probabilistic model to generate new populations, repeat it and achieve the optimization of the population(3-6). Second-order EDA based on general structure Gauss network is introduced to replace the condition probability density function, the crossover and mutation operators are added to speed the evolution process. Combining the individual energy-entropy selection, the detail human-machine interaction forces identification method using the improved estimation of distribution algorithm is given.The quasi-Newton iterative learning control simulation results show the validity of the method.