Pilot Studies on Avrora Unior Car-Like Robot Control Using Gestures

Gesture recognition is not only an important communication channel in human-human interaction but it also allows a human to communicate with other intelligent devices. This paper presents a concept for controlling the car-like robot Avrora Unior locomotion using gestures. We created a list of 18 control commands that contains basic and compound commands. A group of 17 volunteers used this list to create individual control gestures independently. A small part of the obtained dataset of gestures was used with the Teachable machine service in order to preliminary evaluate a possibility of constructing a full-scale model and to train it appropriately. The obtained model demonstrated acceptable recognition rate. We also attempted to apply SURF and FLANN techniques for matching with the direct matching approach and the skeleton-based approach, but the matching results were not satisfactory.