It’s a sad truth, but right now the world is littered with an estimated 110 million land-mines. Clearing them all could take as long as 1000 years and cost $30 billion, but leaving them in situ is not an option. The number of people killed or injured by these hidden weapons recently reached a ten-year high – so how amazing would it be if the Raspberry Pi could help tackle this ever-present problem?
This article was written by David Crookes and appears in The MagPi #63.
Scientists at Arizona State University have been putting their heads together to do just that. They have devised the C-Turtle, a cardboard robot with turtle flippers which has a Raspberry Pi at its heart. It uses machine learning to figure out how to walk across the most unusual and hazardous of terrain, constantly adapting to its surroundings. Modelled on a sea turtle (hence the name), it is not only inexpensive, but easy to transport.
“We were looking to develop a cheap and simple robot for the detection of land-mines,” says PhD student Kevin S Luck, who has worked on the project with Joseph Campbell and Michael A Jansen. “Undetected land-mines are a problem in many countries, and often these mines are particularly difficult to detect in sandy environments. The problem is that sand in a desert moves over time and so the location and depth of the land-mines is constantly shifting.”
The C-Turtle is well equipped to cope with this issue. Housed within a single-sheet laminate comprised of layers of paper, foil and adhesive, it mimics the movement of a sea turtle. The scientific trio had noted how quickly sea turtle hatchlings can move over sand and how adults crawl while lifting their immense weight. This led to Michael developing a workable fin shape, and Kevin and Joseph figuring out how the Pi could best power the robot.
“We envisioned a system where each robot can carry sensors to detect and mark land-mines, but also where the loss of a single robot is relatively inconsequential for demining operations, thus reducing the risk for humans or bigger demining robots,” explains Kevin. During the design process, some key decisions were made. They ruled out using wheels – “they usually have issues with slippage on sand, and they would create a more complex manufacturing process,” says Kevin – and were unanimous in wanting to use a Raspberry Pi Zero.
“The Pi felt perfect,” Kevin continues. “We not only wanted the ability to send commands to the robot via WLAN, but also to perform simple data processing and machine learning directly on the robot – a requirement for using multiple robots in a fully autonomous fleet. The Zero also requires relatively little power. Because of that, we’re exploring the possibility of using solar panels for recharging batteries during the daytime.”
Kevin and Joseph have worked on an algorithm which allows the turtle bot to adapt its crawling technique. “The whole code infrastructure on the turtle robot, from motor control to the joint server and sensor collections, was written in Python,” Kevin reveals. “We used TCP/IP connections to send joint commands to the robot and also to collect data for evaluation.”
This was put to the test when they drove out into the desert with their first prototypes. “We got a real-time feed of what was happening with our robot, and were able to test and debug different variations of the learning scenarios,” Kevin tells us. By using trial-and-error learning, the robot gets good and bad feedback which enables it to develop.
Through this process, the robot has managed to work out effective trajectories over poppy seeds as well as sand, but the scientists are continuing to refine the technology and their ambitions remain high. “We’d like to take the robots into space, too,” says Kevin. “It would be fantastic to use them to explore Mars.”