Where is the gas leak in this complex factory? In the future, flying snoopers could relieve people of the dangerous search: Researchers have developed tiny drones that can jointly locate gas sources in confusing spaces. Your skills are based on bio-inspired artificial intelligence. The concept could also be used in other applications, say the developers.
Extensive facilities with intricate pipe systems, narrow spaces and many obstacles: industrial sites are often highly complex – finding a gas leak on a setting wheel or connecting piece is a tricky task. So far, people have had to search for it with detection devices. Locating the gas source can be time consuming and potentially dangerous. However, letting a robot do this job has previously been difficult. Because artificial sensors are difficult to recognize changes in the gas concentration. In addition, previous robots or drones with the appropriate equipment were comparatively large, which limited their use in complex environments.
Alternative: several small units
So the researchers led by Guido de Croon from the Technical University of Delft came up with the idea of using many small units instead of a single, autonomous search device that work together. “We are convinced that swarms of tiny drones are a promising way for the autonomous localization of gas sources. Because their small size enables them to fly in tight spaces and explore complex environments in three dimensions, ”says de Croon. In addition, the use of the differently positioned units makes it easier to record concentration gradients in order to ultimately detect the gas source, say the scientists.
The challenge in developing their sniffer swarm drones, which are just twelve centimeters wide and 37.5 grams heavy, was to equip them with all the necessary components and artificial intelligence for autonomous navigation and communication. The limitations excluded systems such as those used in self-driving cars. Instead, the scientists sought inspiration from nature.
“In fact, there are numerous biological examples of successful navigation and odor source localization with severely limited capacities,” says co-author Bart Duisterhof. “Just think of how fruit flies with their tiny brains of only around 100,000 neurons pinpoint the location of the bananas in your kitchen. Although we couldn’t directly copy these behaviors, we taught them similarly simple behaviors to get the job done, ”explains Duisterhof. In addition to the search behavior of insects, the scientists were also inspired by the behavior of flocks of birds when designing their system.
Bio-inspired artificial intelligence
As they explain, the basic behavior of the drones is determined by a special algorithm that they call the “Sniffy Bug”. As long as none of the units has sensed gas, the quadrocopter spread out as far as possible over the environment, avoiding obstacles and one another. If one of the drones then perceives gas at its location, however, another algorithm takes care of the collective behavior: The drones communicate with each other and coordinate in order to ultimately systematically encircle the gas source. A single drone only needs to measure the respective gas concentration, but not the gas concentration gradient or the wind direction.
Through tests, the scientists have already shown that their swarms of drones can successfully locate gas leaks in complex environments. However, de Croon and his colleagues emphasize that there is still a need for optimization before a functional product. They are now working on further improving the navigation skills of their little snoopers in order to make them fit for use in real emergency scenarios.
They also see a broader application potential for their system: the algorithms developed could not only be useful for detecting gas leaks in buildings, but also for drones for the early detection of diseases or pests in greenhouses, for example. Appropriately equipped swarms of drones could even be used when exploring Mars – for example to locate methane sources. “Our work shows that swarms of tiny drones can perform very complex tasks,” says de Croon. “We hope that this work will inspire other robotics researchers to include this type of artificial intelligence in their developments in the field of autonomous flying.”
Source: TU Delft