Planning algorithms for robot teams can be centralized, in which a single computer makes decisions for the whole team, or decentralized, in which each robot makes its own decisions. The latter approach is much better in terms of incorporating local observations, but it's also much trickier, since each robot must essentially guess what the others are going to do.
MIT's new algorithm takes a decentralized approach and factors in not just stationary obstacles but also moving ones. Each robot uses its own observations to map out an obstacle-free region in its immediate environment. It then passes that map to its nearest neighbors. When a robot receives a map from a neighbor, it calculates the intersection of that map with its own and passes that on to other neighbors.
Because each robot communicates only with its close neighbors, the bandwidth required for communications is greatly reduced, particularly when there are a lot of robots. And each robot ends up with a map that reflects all of the obstacles detected by the whole team.
The algorithm accounts for obstacles in motion by including time as a fourth mapping dimension. With that dimension included, it describes how a three-dimensional map would have to change to accommodate the obstacle’s change of location over a span of a few seconds.
In simulations involving squadrons of mini helicopters, the algorithm came up with the same flight plans that a centralized version did but allowed for small variations as conditions required.
“It’s a really exciting result, because it combines so many challenging goals,” said Daniela Rus, a professor in MIT’s Department of Electrical Engineering and Computer Science and director of the Computer Science and Artificial Intelligence Laboratory.
“Your group of robots has a local goal, which is to stay in formation, and a global goal, which is where they want to go or the trajectory along which you want them to move," Rus explained. "You allow them to operate in a world with static obstacles but also unexpected dynamic obstacles, and you have a guarantee that they are going to retain their local and global objectives.”
Each robot updates its map several times per second, calculating the trajectory that will maximize both local and global objectives.
To simulate environments in which humans and robots work together, the researchers are also testing a version of their algorithm on wheeled robots whose goal is to collectively carry an object across a room where humans are also moving.
They'll present their algorithm next month at the International Conference on Robotics and Automation.