In the near future, inaccessible environments may be explored by a swarm of marbles powered by artificial intelligence. The work being done in Professor Baltus’ lab at the Eindhoven University of Technology has applications in mapping difficult to inspect areas such as drinking water supply systems or oil pipelines.
Each marble is approximately golf ball sized and tightly packed with circuit boards that collect data from a thermometer, pressure gauge and speedometer. A group of these balls is dropped into the beginning of the pipeline system, after which they are allowed to float freely until they come out the other side and are collected. Parameters of interest are stored locally within the balls — communication with the outside world is generally infeasible or impossible while exploring the pipeline.
Communication between nodes is achieved using ultrasonic waves, similar to the echolocation used by bats and dolphins. While effective, the supporting hardware has a hefty power budget — it consumes five thousand times more power than all other sensors and systems combined. This is problematic when the marbles need to be particularly small, so the use of magnetism is also being explored for communication between nodes. Magnetism has the disadvantage of being effective over only short distances, however, this may be acceptable in a swarm where other nodes are typically nearby.
A swarm of relatively simple sensors is well suited to exploring pipes that are narrow. Packing in cameras, lights, motors — and the requisite batteries — would render a device too large to pass through the pipeline in many cases. Another advantage of the design is that the system being explored need not be shut down or emptied during mapping, which can be very costly.
Before the marbles can be used to inspect a system, they need to be fine-tuned for the task. The balls are sent through the pipeline for an initial pass, and the data collected is used to create an initial model, which is copied to a virtual environment. An algorithm, which was developed with the help of artificial intelligence, then iterates through the sensor readings within the virtual environment to find an optimal configuration for the software that is loaded in the marbles. When this process is complete, the devices are ready to carry out inspections of the system.
Looking forward, the team has hopes that their idea can be applied to even more inhospitable environments — including the inside of volcanoes and the human digestive tract.