Roboswarm is a European research consortium with partners from universities and industry.

The goal of the project is to develop simple and low-cost technologies for making both single robots and groups of robots – swarms – more intelligent.

Swarm applications range from simple cleaning tasks to exploration of large unknown areas, surveillance, rescue, coordinated weight lifting, minesweeping etc. where intervention from human operators is minimized.

Our project uses cheap RFID chips for marking objects like chairs, walls, doors. This is significantly cheaper and more flexible than, say, using cameras on robots for object recognition. The same RFID chips on objects are then used by the robots to leave messages to other robots. The solution is inspired by ants’ communication using pheromone trace known as stigmergy.

The demonstration focuses on making cleaning more efficient. We use a popular iRobot Roomba cleaning robot and attach a small linux computer and an RFID reader/writer on the Roomba, taking over control of the Roomba.

While a standard Roomba is fairly simple-minded, will happily clean places recently cleaned and does not understand that some places should be avoided – or vice versa, cleaned extra often – our system adds necessary intelligence.

First, Roombas understand object descriptions and simple messages written on RFID chips by humans: like „go away”, „fragile”, „clean here”, „this is a chair” etc. When the robot notices an RFID chip ahead, it will read its content and behave accordingly, following the configurable rules on board. The rule engine uses ontologies and allows the Roomba to understand, for example, that chair is a furniture and you can probably go around furniture.

Second, Roombas write their own messages on RFID chips. When the robot notices an RFID tag while cleaning, it will write on the tag that it was cleaning there at that particular time. Next time when it comes near the same tag, it will not clean the place, unless enough time has passed. What is more important, when we use a whole swarm of Roombas for cleaning, all the other Roombas will also avoid cleaning on this marked up place for some time, avoiding wasted work.

This kind of message writing on RFID tags allows the swarm to self-organize without any central planning. For example, if we have a number of rooms in need of a cleaning, say, a whole floor of a building, robots will spread out evenly into different rooms.

The internal database and rule systems we have built use ideas and technologies normally employed in a seemingly quite different area: the semantic web, where the goal is to make web and network software systems understand each others data. We see the roboswarm to be almost like a system of web agents, with different agents understanding what others are doing. All the data and rules are also sent from the robots via a local wireless network to the central database of a robot swarm, which does not coordinate the robot actions directly, but is capable of learning new behaviours and rules for action. These behaviours and rules are then made available to other robot swarms via ordinary web services. In this way we will perhaps one day extend our research prototype to a real world of information-sharing, intelligent robot swarms.