Object Recognition for Humanoid Robots Using Full Hand Tactile Sensor
Somchai Pohtongkam, Jakkree Srinonchat
Abstract
With the development of human-like robotic technologies, robots have been advanced to sense and recognize objects, which is one branch of artificial intelligence to mimic human reactions. Although high-performance computer technologies are used, the ability of object recognition by touch is still low due to the lack of proper sensors. Therefore, this research studied the object learning and recognition system through robot touches by developing an artificial sensory system acting as an electronic skin with tactile sensors. The Tactile Sensor is developed in this research, consisting of 15 Tactile Sensor Arrays and the palm’s touchpoints. Furthermore, recognition analysis was developed on Bag of Word (BoW) and Convolution Neural Network (CNN) algorithms. With the BoW technique, using Support Vector Machine (SVM) as a classifier with Moment Analysis Descriptor (MA) provided the highest accuracy, showing more than 80.15% accuracy from five grasping of an object. With the CNN approach, InceptionNetV3 provided the highest accuracy of 98.28% from only one capture of an object.