Experimental network performance evaluation for human-robot interaction collision detection using cameras

• Autor(a) principal: Oliveira Filho, Assis T. de
• Data de Publicação: 2022
• Outros Autores: Barbosa, Gibson, Rodrigues, Iago Richard, Cani, Carolina, Kelner, Judith, Sadok, Djamel, Souza, Ricardo
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Research, Society and Development
• Texto Completo: https://rsdjournal.org/index.php/rsd/article/view/30543
• topic: AMQP; MQTT; Interacción entre humanos y robots; Comunicación de Red; Enseñanza.; AMQP; MQTT; Interação Humano Robô; Comunicação de Rede; Ensino.; AMQP; MQTT; Network communication; Human Robot Interaction; Teaching.

A practical solution to human-robot collision detection using devices commonly found in workplaces, such as 2D cameras, requires thorough planning and evaluation of network restrictions that may deny timely access to and process important context data collected by IoT devices. In this study, we evaluate the behavior of the AMQP and MQTT application protocols for camera image transmission. We examine the packet overhead for each protocol when streaming video signals. Also, we evaluate the impact of transmission delay on the total decision time starting from the moment the camera captures an image to the moment it is decided if there is a robot-human collision or not. Finally, we also evaluated an open-source platform to emulate wireless Mininet-wifi, seeking to understand the level of influence it would have on the results. The results show that transmission overhead represents as much as 80% of the total decision time and that the AMQP protocol takes around 5% less transmission time than MQTT. The results also show that the use of hardware accelerators such as a GPU increases by 37 times the number of detections. We found that the size of the image to be transmitted and wireless communications did not influence the results for our scenario. In addition, we also noticed that the use of emulation through Mininet-wifi does not negatively influence the behavior of the experiments.