Cooperative running of small robots using character recognition and a robot distance estimation model based on image machine learning

This paper proposes a tracking algorithm and driving control method that can be implemented on edge devices such as microcontrollers, and to achieve practical tracking control on actual robots. To this end, we derive an inference model capable of real-time object detection based on the lightweight machine learning algorithm FOMO (Faster Objects, More Objects). We implement this model on a small mobile robot and propose a multi-robot tracking control method using character recognition based on image learning. Verification results showed that a good inference model was generated, even though the training process used only 136 images in the dataset. This was achieved by using simple images of characters (alphabet) as training data, and a processing speed of 4Hz was achieved on the low-spec microcontroller used in the robot. Next, in a multi-robot tracking experiment, we propose a method for determining the distance and position of a leading robot using character information detected by the inference model, and conducted a tracking experiment using this method. As a result, we were able to achieve both straight-line and meandering following driving with two and three robots, demonstrating the usefulness of the simple tracking control method proposed in this paper. In particular, this control method assumes that all robots participating in cooperative driving share speed information over a WiFi network, making it suitable for edge devices such as microcontrollers, which are being rapidly developed. In addition, in driving experiments, we achieved stable following operation on meandering trajectories by introducing a variable gain for the distance between the robots. In particular, by introducing the relative difference in the posture angles of each robot, the accuracy of estimating the distance between the robots, which is obtained from identification character information, was improved. As a result, we succeeded in achieving more robust following driving using the distance between the robots as a control variable.