Research On Key Technology Of Simulation And Virtual Teaching For Humanoid Robot

With the further improvement of our living standard and the increasing serious of aging problem, we are urgently hope for a kind of Intelligent Robot, which can be brought into our home to look after the more and more aged. Due to the similarity in appearance and behavior with people , humanoid robot is easy to be accepted and accustom to various human facilities. Thus, humanoid robot, as the first choice of robot planed to be brought into human's daily life, is abstracting more and more attention from scientific researchers and becoming a research hotspot in robot.Centering on some key technologies referred in digital simulation and virtual teaching of human robot, based on the analysis on the research status of humanoid robot at home and abroad, this paper mainly studies the digital simulation of human robot under the effect of gravity field. According to the current difficulties in obtaining pre-planning data from behavior panning in humanoid robot, as well as the problems of expensive cost, we proposed a method to realize virtual teaching. We obtained the pre-planning data in virtual manner, and then researched on the modeling of motion mapping of human hand to dexterous robotic hand. The main working achievements and innovative points are as follows:According to the current simulation experiment environment, we applied OpenGL graphic library to explore a 3D real-time simulation system for humanoid robot and in the system we can realize all azimuth observation. Through analyzing the force condition and supporting state under the condition of gravity field, in modes of point support, line support and surface support, we put forward a solving method for the balance condition, the axis of rotation rotated when unstable, as well as the rotating speed of humanoid robot, which can provide evidence for simulating the balance effects of walking robot. In addition, we apply the above method into the 3D real-time system of humanoid robot, and then realized the virtualization and visualization of humanoid robot controlling effects, which can provide a strong research technique for the future research on humanoid robotAccording to the pre-planning problems in controlling the behavior of humanoid robot, we proposed a virtual teaching method based on human robot, and put forward an inverse kinematics method of computing robot arms with seven redundant DOFs and robot leg with Six DOFs. Regarding to the large fluctuating problems of joint velocity and acceleration caused by some reasons during virtual teaching, such as the hands thrilling of operators, we used trajectory planning method based on cubic polynomial to smooth the teaching motion and reduce the influence of the seriously fluctuates of joint velocity and acceleration on motor. Finally, we applied the above methods into the 3D real-time virtual teaching system, and approved the validity and utility of the virtual teaching system through the humanoid robot simulation experiment on the teaching of biped walking, traffic commanding and dancing .Due to structure differences between dexterous robotic hand and human hand, we need to model the mapping of dexterous robotic hand to human hand in master-slave teaching of dexterous robotic hand. According to the lacking universality problem existing in the traditional motion mapping method based on some special dexterous robotic hand designing, we put forward a neural network ensemble training method based on improved K-Mean algorithm, and applied the method into modeling the motion mapping of dexterous robotic hand to human hand. Regarding to the problems that the optimal following posture is hard to obtain when modeling motion mapping, we proposed an inverse kinematics resolving algorithm based on artificial potential field method, and applied the method into sample generation of master-slave teaching between dexterous robotic hand and human hand. Finally, through the contrast experiment on mapping dataset and the master-slave following experiment between dexterous robotic hand and human hand, we verified the rationality and the effectiveness of the method.In order to improve the prediction accuracy of neural network assemble, reduce the error of dexterous robotic hand when following human hand, we proposed an improved Cloud-Adaboosting neural network ensemble training method based on multidimensional cloud. In the method, we regard the individual sample set as an uncertain cloud and make up the insufficient of effective samples problems in the training set by generating drops with the normal cloud generator and backward cloud generator. Then, we modified the outputting weight of individual neural network in the neural network ensemble by computing certainty degree of test samples towards the individual training samples. The contrast experiment results show that the method can furtherly improve the prediction accuracy of neural network ensemble, which play positive role in improving the master-slave following effect of dexterous robotic hand towards human hands and enhancing the efficiency of teaching.