Amazon may have just gotten its claws into Kiva Systems, but there's more than one company out there looking to automate warehouses with smart little robots. At the Fraunhofer Institute for Material Flow and Logistics, researchers are looking for ways to make warehouse robots smarter and more efficient by getting them to communicate and cooperate like a swarm of ants.
A swarm is just exactly what you want with warehouse robots. There are a lot of them, and they're all identical and interchangeable, cooperating to complete complex tasks by combining simple actions. The big difference between a swarm of (say) ants and a swarm of (say) robots is that the ants don't have any high-level control: each ant has its own little tiny brain, and even though ants have specific tasks that they are directed (or bred) to perform, they decide on an individual level how to go about carrying out their instructions.
What Fraunhofer is trying to do is mimic the ant swarm system with robots. For example, instead of having one central computer control the movements every robot (as with Kiva), Fraunhofer's system utilizes robots that make their own decisions with onboard computers. Each robot communicates with all the other robots in the swarm simultaneously using WLAN, and they use algorithms based on a model for how ants forage for food to cooperatively decide which of them should go where and do what.
The robots don't need fixed localization points, but instead rely on "integrated localization and navigation technology" (including signal-based location capability, distance and acceleration sensors and laser scanners) to find the most direct routes to their destination without crashing into anything or each other. This makes them very efficient, and it also makes the system easily scalable, since you can introduce new things and the robots won't freak out.
Scalability, reliability, and flexibility are why swarm robotics has been getting so much attention lately: need a bigger system? Just toss more bots into the mix. Lose a bot to a mechanical problem? It's not a problem, since another bot just takes over. We've seen lots of swarms related to search and rescue (i.e. military) applications, but as far as a way to improve a commercial (or industrial) project, this research seems like a promising way to go.