Food Delivery Planning Of Meal Assistance Robot Based On Visual Recognition

With the development of computer technology and artificial intelligence,robot technology continues to integrate the latest scientific and technological achievements,and develops in the direction of informatization,intellectualization and multi-function.Nowadays,as one of the effective means to solve the nursing problems of the elderly and the disabled,the field of service robots has attracted widespread attention from the scientific community.In order to solve the problem of visual servo of the meal assistance robot and improve the intelligence and humanization of the robot,the visual servo control method of the meal assistance robot based on artificial intelligence is studied and experimented in this paper,which provides a solution for the visual detection and motion planning of the robot and has practical significance.Based on deep learning technology and reinforcement learning technology,a detectionplanning scheme is proposed,which can complete the motion planning of the whole process from picking up food to delivering food to the user’s mouth.The diagram of detection-planning program is established,and the hardware selection and software development environment are completed.Based on the task requirements of the meal assistance robot,the characteristics and advantages of the overall method are analyzed.In this paper,I build and train a face detection model based on yolov5 object detection algorithm and lightweight convolutional neural network,which can capture the key points of the user’s face in the input video stream,and use binocular stereo vision technology to obtain the three-dimensional coordinates of the key points.By processing the facial key point data,this paper uses the way of establishing the head coordinate system to weaken the influence of occlusion and other problems that may occur in the process of food delivery on the measurement accuracy of binocular stereo vision.Finally,a measurement system which can obtain the user’s mouth position in real time is established,an experimental platform is built,and the measurement accuracy experiment is completed.With the data support provided by the detection model,this paper builds an agent for food delivery motion planning based on reinforcement learning,builds the value network and action network based on the idea of actor-critic,and analyzes and deduces the state space of the planning environment and the action space of the agent.Based on the requirements of rapidity,stability,accuracy and real-time of the food delivery task,this paper builds a reward function to guide the direction of the agent network update.Finally,the training of the agent is completed by using the computer to simulate the movement of target point.In this paper,an experimental platform is built and the motion planning experiment of the meal assistance robot is completed.Experiments show that the detection-planning method can complete the food delivery task according to the expected task goal in the two cases of multidirection motion and natural motion of the target point,which confirms the accuracy and effectiveness of the method.