Research On Tactile Recognition And Speech Interaction Of Robots Based On Learning Algorithms

With the rapid development of intelligent robot technology,its applications have spread on many fields,such as the medical,agricultural,scientific research,education,entertainment,industrial,military,aerospace,etc.The relationship between robots and humans has become increasingly close.Thus,a safe and effective interaction between robots and the external environment has become more and more important.On the other hand,how robots provide services to human beings in a natural and fluid way has gradually become an important research direction.As an important form of perception,tactile sense is able to help robots to realize the perceptual understanding of the external environment.As the most common and convenient communication method in human daily life,speech dialogue has gradually been used as an important tool for intelligent robots to communicate with humans.On the one hand,this paper develops an algorithm based on extreme learning machine(ELM)for uncertain robot manipulators,which can detect an unknown object and estimate its elasticity and boundary.On the other hand,we build a voice dialogue system based on the open source framework,so that the robot has the ability of voice interaction with humans.The user can control the robot to provide tactile detection,moving objects,weather query,instant chat and other services.The specific research work of this paper is as follows:1)Design a tactile recognition controller for a robotic arm with dynamic uncertainty.In this paper,an ideal reference model is set as the target performance index of the closed-loop system.ELM is used to compensate the uncertainty of the dynamic model of the manipulator,so that the dynamic closed-loop system of the manipulator can track the matching reference model.By adjusting the feedforward force and reference point in the reference model iteratively,the stable equilibrium state reached when the robot manipulator interacts with the object is recorded,and thus the elastic coefficient and geometric boundary of the detected object are estimated by weighted least squares method.2)Design a novel learning law for the update of ELM parameter.In order to make the dynamic closed-loop system of the manipulator track and match the specified reference model in finite time,we design a novel learning law for the update of the output weight of ELM,and proves the effectiveness of the algorithm through Lyapunov.The simulation experiments on MATLAB demonstrate that the ELM-based controller proposed in this paper can make the dynamic closed-loop system successfully match the specified reference model in finite time,and can estimate the elastic coefficient and geometric boundary position of the object through iterative detection.3)Build an intelligent voice dialogue system to enrich the conscious form of the robot.Based on Rasa framework,we build a dialogue platform for robots to realize natural language understanding and session management/behavior decision-making,and encapsulates it into API form to provide services.The API services on the XUNFEI AIUI platform for automatic speech recognition(ASR)and text-to-speech(TTS)are integrated into the above dialogue system to realize the voice input and voice output functions.In addition,the Chat Robot API provided by Turing Robotics is used to implement the chat function for the dialogue system.The voice dialogue platform built in this paper enables users to control the robot to provide tactile detection,moving objects,weather query and other services through voice,which enriches the diversity of human-computer interaction.