Working on the research question “Could robots soon help rescue crews save the survivors of a natural disaster?”, scientists from Université libre de Bruxelles undertook a series of trials. For group-action by robots to be a success, the machines would need to be able to determine which tasks to perform and the order that the tasks need to be performed in. These activities also need to happen without human intervention. The scientists showed how this ability can emerge from a group of robots more effectively than from robots working by themselves.
As an example of the need for robots wok ring together to reason is with emergency responses to natural disasters. For example, if a robot detects a survivor there is little use the robot seeking to pull a victim out of rubble if the rubble itself has not been remove. Currently, technologists need to instruct robots what to do. The new research provides an alternative.
This involved modelling swarm robotics, drawing upon from the collective behavior of social animals (like ants). This was used as the basis to develop a set of robots with the necessary artificial intelligence. This moved beyond robots that are able to communicate and coordinate to make decisions and perform simple tasks, to producing a swarm of robots that can undertake a sequence of three actions, without knowing the correct order in advance.
In trials the test robots were set the task of moving to three different points in space, where they were instructed to perform a simple task. Only after the tasks were completed would the robots learn whether the order was correct. Hence trial actions became the basis of machine learning. While further study is needed, the research paves the way for developing a new generation of smart robots.
The new study has been published in the journal Science Robotics. The research paper is headed “Autonomous task sequencing in a robot swarm.”
