Type-2 Fuzzy Logic Induced Gesture Recognition For Robot Task Planning

With the advancement of robotics in industry and healthcare, a key component of research is to develop sophisticated design automations to offer precise and accurate response of the system with the freedom to access the system easily. The paper proposes a novel arm-gesture-driven position control of a robot arm, which can be operated by any person even without specialized domain knowledge. The uncertainty in gestures is modeled by type-2 fuzzy membership functions. Type-2 fuzzy rules are employed to recognize gestures. The robot learns different tasks from human demonstrations, and finally principles of non-monotonic reinforcement learning are used for mapping the gestures to the tasks. Over 20,000 gestures obtained from 10 subjects are used to validate the effectiveness of the gesture recognition scheme. One interesting parameter is employed in the learning dynamics to control the unlearning rate of old knowledge and learning rate of the new knowledge about gesture-to-action mapping in non-monotonic circumstances. The choice of this parameter is left to the user for his specific application. Experiments have been undertaken to confirm that the proposed system works successfully with negligible errors in angular displacements of all the joints of the robot arm. The proposed system is expected to have applications in robotic surgery and task-planning in hazardous environment.