Planning algorithms for robot teams can be centralized where each computer makes decisions for the entire team or decentralized, where each robot makes his own decisions. The latter approach is much better in terms of incorporating observations of locations, but it is also much more challenging since each robot must essentially guess what others are doing. The new MIT algorithm assumes a decentralized approach and factors that not only fixed obstacles such as those in motion. Each robot uses its own observations to measure a region free of obstacles in its immediate context. Then it passes such a map to its nearest neighbors. When a robot receives a map from a neighbor, it calculates the map of the intersection with its own and follow the changes to their peers. Since each robot communicates only with its immediate neighbors, the band rate required for communication is reduced significantly, particularly when there are many robots. And each robot ends up with a map that reflects all the obstacles detected by the entire team. The algorithm takes into account moving obstacles to include the time as a fourth dimension in the mapping. This dimension describes how a three-dimensional map should be changed to accommodate the change of location of the obstacle in a range of seconds. In simulations with drones, the algorithm achieved the same flight plan that a centralized version has, but allowed small variations as they were required. Daniela L. Rus, a professor in the Department of Electrical Engineering and computer scientist and director of the Computer Science and Artificial Intelligence Laboratory said that “It’s a really exciting result by combining many challenging objectives”. Each robot updates its map several times per second, calculating the trajectory that will maximize both the local and global goals. For similar environments where humans and robots working together, researchers are also testing a version of its algorithm robots equipped with wheels whose goal is to collectively carry an object over a room where humans are also moving. The researchers will present their algorithm next month at the International Conference on Robotics and Automation that takes place in May in Stockholm, Sweden.
Δ
