Google recently unveiled a plan to provide wifi via baloon. This interesting idea is part of an attempt to cover the globe in internet connectivity and fits nicely into a larger plan.
They also bought Boston Dynamics, who makes robots that could be considered military grade, as well as 7 other major robotic companies.
They're are building out a knowledge graph which would allow you to ask questions and get a single, but correct, response.
They're building out conversational search, so that the knowledge graph and it's responses can be accessed through the same 'interface' we use daily in speaking with other people.
They already have cars which can navigate most roads in most conditions without human intervention, so long as they have an internet connection, and does so by mapping the environment.
With that in mind, they're working on a phone like device which would be able to map any physical space into a three dimensional representation of that space.
So they are independently working on mobile robots who can assess and navigate physical space, conversation, and social norms through a nationwide, uninterrupted, high speed network.
Other parts of the equation that don't fit so neatly are Sentiment analysts would allow the bot to understand moods to improve response accuracy. Semantic understanding of vague sentences allowing for more human modes of speech. And a million more studies on machine learning to help robots to program themselves.
We can also look at each of the robotics companies acquired by google to get a different perspective of what's going on.
Schaft - Body/mobility - Winners of the DARPA Robotics Challenge, this team has made a great bipedial robot.
Industrial Perception - Eyes - These guys work on the more classical robot arms you can currently find in industrial plants, though the arms are made special by the fact that they use 3D vision as guidance.
Redwood Robotics - hands - Wants to make, no joke, the Apple II of robotic hands.
Meka Robotics - Body/Image - Their focus is making robots that are safe around people and their speciality is building arms.
Bot & Dolly - Motor control/Image - a lot of people seem to think these guys will provide better control to industrial robots, which makes a lot of sense. I, on the other hand, think these folks were acquired to tackle the PR problem around robots. If Google wants to make consumer bots, they are going to have to do a lot of work to gain consumer trust.
Autofuss - Image - Sister company of Bot & Dolly, they make pretty amazing videos - including a lot of the shots from Gravity.
Holomini - Body/Mobility - They make omnidirectional wheels.
Boston Dynamics - Body/Mobility - The superstar, they crank out 2 and 4 legged bots.
DeepMind - The mind - Google's software is working on a 'brain' of sorts already, deepmind was likely hired to help fit that mind into a robot body.
Titan Aerospace - Land mapping/ Network
Skybox - Land mapping/network
I want to talk about the last two separately as they fit back into the original idea and project loon. Skybox has some neat services for satellite video, real time analytics, and the ability for third parties to give commands to the cameras. Titan Aerospace (whose site is currently blank) makes high flying drones that can stay up in the air for a reported three(!) years. Looking at these in tandem with the ongoing loon project, it seems very feasible that google could turn maps into a real time video service, so long as their flying network has the bandwidth.
The analytics would allow real time traffic reports, but it would also collect a huge amount of practical data on traffic patterns to help route cars efficiently. Of course, that's not where the data would stop as all changes to the landscape - manmade and natural - would be catalogued and analyzed. Real life examples are often charitable, such as emergency response or deforestation monitoring, but it would be easily applicable to everything from zoning to criminal tracking.
Add on top of this the GPS data that transmitted from our phones and within five years you could create a very accurate digital representation of the entire physical world along with the common movement patterns of its inhabitants.
To go much further into what that world would look like I'd have to just start cranking out sci-fi stories.