RACoN: a robot activity recognition approach using a convolutional neural network for the RoboCup Standard Platform League Penalty Shot Challenge

Abstract

In this paper, we present the early stage of the development of a multi-agent system. Embodied inside a Nao robot that will perform in the RoboCup Standard Platform League Penalty Shot Challenge, called RACoN (Robot Activity Convolution Nao). With this approach, the Nao robot will be able to interact with their environment and be aware of the activities of the other players. By mapping cameras, sensors, and actuators as individual agents across the ROS environment. Allows us to integrate then across the RaCoN Architecture, transforming it into a Pandemonium Artificial Cognitive Systems. In this kind of workspace, the robots will detect and recognize other robot activities using a ConvNet (Convolutional Neural Networks). These convolutional neural networks are trained to discriminate between teammates, opponents, archery, the ball, and the field, allowing the kicker and the goalkeeper to perform well complying with the RoboCup standard league rules.