Learning Robotic Manipulation Tasks From Human Motion Via Probabilistic Approach

People use their hands every day to do things like pick up objects, use tools, make gestures, and communicate using sign language. Current robot hands are not able to do things as well as humans. In recent years, one powerful method that has gained prominence in the field is teaching by demonstration.With this method, the robot can learn and copy actions just by seeing them, making it easier for people to teach robots and have them understand us better. The motivation behind this research is to address the challenges posed by teaching by demonstration, particularly in the realm of data acquisition and processing regarding some specific tasks related to grasping and manipulation of objects. The aim is to propose a viable solution by developing communication libraries tailored for the available hardware, namely Cybergloves and a spatial hand tracker. This enables the capture of hand movements and spatial positioning. Subsequently, the research focuses on the crucial step of data processing, employing Hidden Markov Models (HMM) with Gaussian Mixture Regression (GMR)[1]. The combination of hidden Markov model (HMM) and Gaussian mixture regression (GMR) allows us to extract redundancies across multiple demonstrations and build robust models to reproduce the dynamics of the observed movements.The ultimate goal is to establish an effective framework that leverages these technologies to enhance teaching by demonstration, offering a comprehensive solution to the identified issues in data acquisition, processing, and application of machine learning techniques.