When computational or storage demands exceed the on-board capacity of a robot, they are offloaded to the cloud, where the massive resources of a datacenter can supplement their limited local resources. Cloud robotics also represent a significant advance for robot learning. Where it might take one robot 150 hours to learn a task using its own individual artificial intelligence (AI) resources, the collaborative effort of 150 robots learning different parts of a task might complete the task in an hour.
With a Wi-Fi connection to cloud-based resources, a robot can access a vast library of known objects to identify things in its environment. Object recognition helps a robot to better perform tasks like sorting, cleaning and using appliances. Relying on the cloud for resources also means that the robot itself can be simpler, eliminating costly compute power and the associated cooling and electrical power draw. Due to this offloading, cloud connected robots have lower battery requirements and are overall lighter and less expensive. Cloud robotics are typically used for tasks that don't require real-time execution, preserving local resources for applications with demanding time constraints.
The connection to the cloud eliminates the need for a robot to learn a task any other connected robot has: It can download the necessary information instead of having to feel out or observe how to do a task. This interconnectedness can help robots work together more smoothly too, coordinating their tasks automatically.
Google’s self-driving cars are one type of cloud-connected robot. The autonomous cars access data from Google Maps and images stored in the cloud to recognize their surroundings. They also gather information about road and traffic conditions and send that information back to the cloud.
Another example is Romo, an inexpensive ($150) cloud-connectable robot that moves on a treaded base housing a battery. An iPhone provides computational power and its camera, microphone and speakers. The cloud compute factor makes the robot better able to recognize its environment. Romos can be used to play games or can be used for telepresence, which allows a human operator to be virtually present in a remote location.
See Ken Goldberg's lecture on cloud robotics: