Testing the possibilities of artificial intelligence
January 01, 2017
|Dr. Forrest Stonedahl with students testing robots over the summer.|
As computers become more powerful, what does that mean? Are they intelligent? Are they a threat? Such questions guided the first artificial intelligence class at Augustana College, taught by Dr. Forrest Stonedahl.
With the help of a new faculty research grant, he bought LEGO Mindstorms robotics sets to incorporate in his research, teaching and extracurricular activities. Dr. Stonedahl sees potential to use swarm robotics after natural disasters and has imagined a team of robots locating, assisting and providing a beacon to rescuers.
So, with a team of his students and a collaborator from St. Ambrose University, he assembled, programmed and tested the LEGO robots' ability to work cooperatively in a search-and-rescue-inspired challenge.
For the initial research, a U-shaped maze was constructed, and the robots were tasked to find markers on the floor. Once a robot discovered one, it would beep and stay with it while the others kept searching. Dr. Stonedahl endowed these silicon-based automatons with artificial neural networks, which he refers to as "robot brains."
Using a custom software robotics simulator that he wrote, he used the principles of Darwinian evolution to "evolve" whole populations of robot brains over hundreds of simulated generations. This project tested whether more effective robot intelligence could be evolved by purposefully seeking and rewarding novel behavior that was different from previous robots, versus rewarding robots for direct progress toward the task objective.
The robot brains were downloaded into the physical robots and tested in the actual course. The results of the experiment closely matched the simulated results, and the evolved robot brains outperformed a program that humans had written for the robots, marking another small triumph for AI over human ingenuity.
However, Dr. Stonedahl noted that the road to robotics research is littered with challenges, including the machines' propensity for breaking down, not following directions to a T, and other real world challenges that simply don't come up in simulation.
After wrapping up the research project, Dr. Stonedahl used the same robots in his AI course. The students' final project was to work in pairs to program the robots to use a technique called backtracking to find a goal marker hidden inside a walled labyrinth, then return to the starting location and display a map of the labyrinth that it just learned.
Invitational Robotics Challenge
His work culminated last spring when he coordinated the Augustana Invitational Robotics Challenge, an extracurricular robot contest open to students across the college. This interdisciplinary event, which engaged students from physics, engineering, computer science, and other fields, involved both building and programming a robot that could quickly climb over a series of hurdles, keep track of its progress, turn around and return to the starting line.
Six teams vied for the win in a competitive and friendly atmosphere. It was a successful venture that Dr. Stonedahl hopes will become a tradition.
While the robots showcase a physical side of computer science, Dr. Stonedahl's broader research includes studying complex systems, simulated systems, and the philosophy of "emergence" - that is, the evolution of complex behaviors from simple rules, and finding interesting patterns in between the realms of order and chaos.
Further possible research applications include the scientific modeling of natural systems, collaborative problem-solving involving multiple agents, and automated analysis of large data sets, which can mimic computer programs, build games, and track the evolution of AI.