Search and Rescue with Autonomous Robot Teams
A fugitive on the loose may not just be fleeing from U.S. Marshalls but a fleet of autonomous robots. Researchers at the University of Michigan are developing the protocols needed to empower individual robots to conduct thought experiments as they make decisions on how to proceed during their mission.
“We take inspiration from humans,” said Edwin Olson, the director of the APRIL Robotics Laboratory in the computer science department at UMich. “When humans work together, they don’t share every irrelevant piece of information they get—they decide what information is actually useful to others.”
Previous attempts to use autonomous robot teams to complete a task often assume unlimited, reliable, and instantaneous communication across the team. According to Olson, this expectation is unrealistic. To counter communication limitations, his team is developing protocols that each robot can use to evaluate the next best step during their mission.
“The decision making process is decentralized,” said Max Krogius, a doctoral student in Olson’s lab. “Each robot is making its own decisions using information that comes to it and is predicting how useful that information would be to its teammates .”
The team is building on the Multi Policy Decision Making approach, which allows the robots to flip through different team strategies in their library as they search, explore, and navigate their world. Olson and his team are testing the flexibility and resilience by having the robots play games.
“We play hide and seek with a team of robots where we can control the character that is trying to hide,” said Krogius. “When the robots are using one policy, it is easy to hide from them, but when the robots use the decentralized multi-policy algorithm, the robots always find the evader in the end.”
What was unexpected in this study was how the robots draw on and expand the insights provided by the protocols in their library.
“What is exciting is the robots often produce mash-ups of the protocols in real-time to create something new,” said Acshi Haggenmiller, a doctoral student in Olson’s lab. “These unexpected combinations are not what our research team initially intended, but can be uniquely well suited to the specific scenario.”
The results of this study have military applications, improving search and rescue missions as well as the deploying humanitarian support. It also has civilian applications, such as improving the function of self-driving cars.
“In our robotics lab, we like to combine new algorithmic ideas with real-world systems,” said Olson. “Seeing robots work together, solving problems that matter to real people is incredibly rewarding.”
Olson, Krogius, and Haggenmiller were joined by Denise Rizzo at the U.S. Army Ground Vehicle System Center on the project, titled “Communication-Constrained Multi-Robot Coordination.”
Stacy W. Kish