Machine learning of haptic objects and reproduction for virtual reality

Abstract

umans interact with machines extensively through our auditory and visual senses, but they frequently ignore their most trusted sense: touch. However, the sense of touch has tremendous potential in almost all fields, including medicine, exploration, industrial robots, and gaming. Haptics, or the science of touch, enables humans to not only remove the barriers to achieve realization in virtual world, but also to perform a wide range of real-world manipulation tasks. Unlike visual and auditory senses, sense of touch is bilateral. Thus, realistic haptic feedback takes utmost importance to achieve realization in the virtual world and to enhance human performance in the real world. Prior studies have used an environment model to reproduce the haptics sensation from the environment as if it’s from the real environment. Most studies have employed conventional spring damper model to model the environment model and motion parameters were considered as the factors affecting for force response. However, the traditional spring damper model doesn’t reflect the actual object. Furthermore, the influence of learned force on reproduction requires special consideration, but haptics studies mostly consider motion data. However, there can be several factors affecting the recreation of haptic feedback. Thus, it is essential to analyze these factors to precisely reproduce an object in haptic dimension. Most studies have utilized force/torque sensors despite their shortcomings such as narrow bandwidth, signal noise, complicity, noncollocation,

and instability. However, robust sensorless force/torque control over a wide bandwidth can be achieved using observer techniques and Disturbance Observer (DOB) and Reaction Force Observer (RFOB) are primarily used to get force measurements. AI enables computers to utilize vast quantities of data and employ their acquired intelligence to arrive at optimal conclusions and uncover insights in mere fractions of the time it would take for humans to do the same. Thus, recent technological studies have focused on using AI techniques to analyze larger and more complex data sets to achieve accurate and faster results. Thus, incorporating AI with haptics allows seamless integration with virtual reality and tune this technology to achieve precise responses. Thus, this study focused on introducing machine learning and deep learning based vivid force sensation reproduction through a virtual model which replicates the actual environment. The information needed is abstracted through Disturbance Observer (DOB) and Reaction Force Observer (RFOB) based sensorless approach. Furthermore, statistical analysis was conducted on data to identify important features affecting the target value of force response. Keywords — Haptic interaction, Force response, Disturbance Observer, Virtual reality, force response, motion parameters, Artificial Intelligence, correlation, Principal Components Analysis (PCA), Random Forest, Haptic object Reproduction, RMSE.