Robotics

Like many others I was constantly checking the BBC livefeed and ESA reports yesterday as Philae landed. My interest is twofold (well threefold, but we will get to that):

- Firstly, exploration of space is always fascinating and who knows what we will learn from this (It's quite scary the number of people arguing this is a a waste of money!)
- Secondly, it's another episode in the application of the advances in modern Robotics - and for a good, not a dystopian, outcome. The potential of robotics is well showcased here - the downside (surveillance by Philae's brethren) is more worrying, but can be put aside for now.

This second point has me musing a bit more. It has been clear to us for some time that space exploraton will, in the main, be done by our robots for a long time to come. I'm sure I'm not the only one who is half hoping that Philae spots the carcass of another civilisation's robot sent to look at the comet.

Which brings me to my threefold muse - to wit, its highly likely that other civilisations will also use robots for their space exploration, so the most likely human-alien first contact will probably be between two robots. Not so much a "take me to your leader" moment, but a struggle to find a compatible protocol to communicate with each other (and maybe a race to recharge the batteries in the only available plug socket).

As readers of this blog may know, we started watching Robotics again in about 2008* when it became clear to us that the advance of Moore's Law meant that finally enough computer power and battery life made it possible to build robots with all their systems onboard, so they could become self-mobile and (to an extent) self directing. This change was labelled "3rd Generation" robotics (as always it was hyped long before it became reality, but over the last 5 years or so there has been a tipping point), and heralded a "Cambrian Explosion" in new robot design. This happens in every new technology, see here re: ships, steel and steam for example, and a plethora of ideas (and companies) start up up in the Darwinian ooze of the startup ecosystem, until eventually the category killers emerge.

Anyway, you know the robobuzz is well and truly ringing the bells of the early mass mind when someone like NESTA produces a 100+ page book on Robotics, it is called "Our Work Here is Done", and I read it over the weekend. It's a series of essays by various people on the topics of robot evolution, robot economics, robots and society, and of course What is To Become Of Humanity.

As to the individual papers - as you'd imagine, they're a bit curate's eggy, but there is a lot of good stuff, and some real nuggets in just about every paper. There are 4 main sections:

1. The Economics of a Robot Future (Has a lot of Social impact thrown in)

2. Technology Possibilities

3. Robots of the Past & Future

4. Robots and Justice (which has a whole 'nother slew of Socio-Economics)

Overall though it is a good introductory overview to all the emerging socio-economic issues. My two critiques overall are that:

- it is light on the actual technologies and how it all works, IMO its useful to have a grasp of the basic emergent technologies - stops the flights of the the fanciful.

- there is a bit of Panglossing over the downsides. Who owns these automated means of production for example? To whom will the newly liberated labour productivity wealth flow?. Promises of a leisure-filled and abundant future for a happy humanity should be put in the same bin that the self same predictions in the 1930's and 1970's were.

But with those caveats, its a cracking read.

I'm sorry I missed the launch, as one of Broadstuff Towers' all time heroes, Carlota Perez,, gave a keynote talk and it was excellent (see here). She would be the first to note there is usually quite a bit of Destruction before the happy Creation phase. I suspect that will also be the case for robot futures - take longer to happen, be nastier while happening, and take longer to get better. As one grounded participant (Ben Russell, Curator of Mechanical Engineering at the Science Museum) tweeted:

"The cry of forthcoming robot revolution won't be "I'll be back" or Exterminate, but 'Unexpected item in bagging area"

No one knows what to do with those who will be displaced by this industrial revolution, but at least this time round most of the writers note the displacement will happen...

(*Bit of background - my Honours dissertation & design project was on Robotics, years ago when they were very 1st generation - I reckoned robots were at least 30 years away from being more than auto-Waldos and went on to d other stuff, but now things are getting very interesting again).

From The Atlantic, it would seem that Google self drive cars rely hugely on the information about the road.:

Today, you could not take a Google car, set it down in Akron or Orlando or Oakland and expect it to perform as well as it does in Silicon Valley. Here's why: Google has created a virtual track out of Mountain View.
A Googler demonstrates what the self-driving car "sees." Note the background world on which the light traffic lights and their sightlines are displayed: that's the track.

The key to Google's success has been that these cars aren't forced to process an entire scene from scratch. Instead, their teams travel and map each road that the car will travel. And these are not any old maps. They are not even the rich, road-logic-filled maps of consumer-grade Google Maps.

They're probably best thought of as ultra-precise digitizations of the physical world, all the way down to tiny details like the position and height of every single curb. A normal digital map would show a road intersection; these maps would have a precision measured in inches.

In other words, the real work is not in the software for driving the car, but in mapping the road...

Google has created a virtual world out of the streets their engineers have driven. They pre-load the data for the route into the car's memory before it sets off, so that as it drives, the software knows what to expect.

"Rather than having to figure out what the world looks like and what it means from scratch every time we turn on the software, we tell it what the world is expected to look like when it is empty," Chatham continued. "And then the job of the software is to figure out how the world is different from that expectation. This makes the problem a lot simpler."

While it simplifies the driving algorithms, it relies on a hugely detailed - and up to date - map of the roads being used. In a way it makes me happier - they are not using new super-technology, nor have they thought of anything no one else has - they have just imagined using a far bigger hammer. As The Atlantic points out, the solution is very "Googley" - Mohammed won't go to Mountain? Simple - move Mountain...

The more you think about it, the more the goddamn Googleyness of the thing stands out: the ambition, the scale, and the type of solution they've come up with to this very hard problem. What was a nearly intractable "machine vision" problem, one that would require close to human-level comprehension of streets, has become a much, much easier machine vision problem thanks to a massive, unprecedented, unthinkable amount of data collection.

Not to mention massive use of huge data centres to do the Crunching...

Which is all very well, until you start to look at the costs of producing this sort of data on enough roads to make these sort of cars useable by private motorists. In fact the setup costs are so huge, and any payback very skewed (looking at where most road miles are travelled, it looks a lot like a logistics network) it's likely that these sorts of cars will initially be used as a new sort of taxi in or between major conurbations (ie cost to datamap road can only be offset by volume of paid journeys on very heavily trafficked routes). "Data Highways" first, then city roads, then suburban roads, then small towns, and as for the final 20% of rural roads - maybe never. And that is just the cost of initial digitisation - but you then have to keep those data maps to a useable level of accuracy, so frequent re-digitisation is required.

This also seems very similar to the classic Simulation Model error - the belief that if you can make a model that is accurate enough, you can simulate anything in software. The truth is you cannot simulate anything with 100% error free models, so you have to believe that 99.9999999% reliability x millions of road miles travelled x probability of serious accident will give a better outcome than is currently possible with a zero cost non digitised road, a sophisticated biocomputer driving the vehicle, and a fraction of this investment spent on a fraction of this technology to put intelligence onboard to make safer cars over the next few years.

The final issue is this - if Google spends the $ billions to collect the data, and if no one else can use these roads with their automated cars unless they buy that data (how else will the business case work - Ad boards on the cars? ) - then that is a classic infrastructure lock in, like any good road toll scheme. Of course, the local authorities could then add an extra tax on using these cars on their roads, so this could all get very interesting....

So - I made that Ad comment somewhat snarkily, only to find out that very evening that some people are planning to attach plastic laminates that can have have Ads printed on them on cars...and the laminates can have new images uploaded onto them from time to time.

Saw this project - Outrunner -on Kickstarter. It particularly caught my interest as it is using a simple but innovative approach to robot movement - two rotating 3-spoke "legs" to get in effect a 6 rotating leg device - to move extremely fast (c 20 mph) and uses a combination of weight shifting and speed adjustment to steer.

It's interesting as the search for optimal robot movement systems is in its early days, and the level of experimentation is huge. This approach maps to the cycle of human running movement, but abstracts it to another architecture (rotating spokes) to deliver a similar result - in my view this observation of tested bio-mechanism and abstraction to mechatronic ones is probably going to be the endgame for optimal "mechanical life" configurations - but this field is a proper "artificial Darwinian stew" as robospecies evolve, and who knows what the endgame will be. (There is a 2 x 2 4 legged version as well, interesting as both sides have 4 spokes but 2 are shorter an don't reach the ground - the spoke on the other side does- I assume that is to balance the secondary moments - one wonders therefore why a 2 x 4 arrangement is not better than a 2 x 3)

But the bigger picture is this - this project (and the many like it) are being funded by end customers rather than angels or banks, so control of the money and ownership of the business is far more in the hands of the founders. Compare this to what these people would heve had to do to get the product out even a few years ago. Whether this one wins or loses is irrlevant, many are being tried out, innovation is moving at rapid speed

One of the real lesson we have learned in our research on new technologies that take off, is that the technological innovation is not enough to drive success and that often innovation in the business model is also required. The "Equity Gap" in technology startup funding has been well documented (by us as and by many others), and it seems that these crowdfunding schemes are part of the solution to that gap*. I see it as a part of the emerging collaborative economy, and it is massively disruptive as it:

- cuts out all direct middlemen, and replaces funding with a market platform at far lower rental cost
- gives the entrepreneur very direct customer feedback very early on (no likee, no fundee)
- keeps ownership with the innovator longer

In fact one could argue there is now a Darwinian competition between business models as well as mecatronic perambulation systems.

Of course, this is not to say this sort of funding won't have the same problems every other new player going to market has - getting attention. Apps were cool when there were a few, its far harder to make a splash when there are millions to sort through. No doubt at some point rising marketing spend and "trusted third parties" will interpose themselves between funder and crowdfundee (though they may be bots as much as humans) - but for now, it good to see innovative technology being driven by this new innovative business model.

* As of course is the "Freeconomic" benefit of massive research by semi publically funded bodies that these technologies often spin out from....

Some time ago we showed a man carrying dronecopter, and then the idea of Amazon delivering packages via drones - put that together, add a few more horsepower, and why not a truck-copter? Now you can - from the Torygraph, news of a truck capable of flying with an octocopter arrangement of rotors:

In this first test, the vehicle hovered 10 feet off the ground for a couple of minutes, but it is capable of flying at much higher altitudes while carrying payloads of more than 4000 pounds.

After landing, the UAV folds up its propellers and transforms into a supply truck, capable of navigating over rough terrain.

There's more....why stop with trucks:

The transformer's drive train will also be detachable, allowing it to be replaced by an amphibious boat hull, making it suitable for missions on land – and at sea.

What's interesting to me, putting my Engineering hat on, is why this is only possible now - the theory has been around for decades, so why now? Turns out that its the controllers that allow multiple rotors to be co-ordinated to keep the machine stable, these simple rotors don't tilt or change pich so can only move the 'copter by different motors speeding up or down. Which is the other thin we can do better today - electronic controllers and motors now are better at speeding up and slowing down. Even so, these vehicles will always be less efficient than helicopters. and as vehicle mass increase this will increasingly be a problem so there is probably a limit to growth here - but the (computer aided) simplicity of these new rotor designs are very tempting for smaller man-carrying craft.

Kudos Gangstersout blog

Two pieces of news in quick succession - Friday, drones are cleared for use commecially in the US* - Pando Daily:

Drones might soon be coming to a neighborhood near you. A federal judge has ruled that the Federal Aviation Administration has not explicitly outlawed the unmanned aerial vehicles. This will allow all number of businesses, from beer distributors to photography companies, to operate their drones without running afoul of the law.

The ruling was made in favor of Raphael Pirker, a drone pilot fined by the FAA for “operating a drone recklessly while filming at the University of Virginia,” according to Politico. Though the FAA could issue an emergency rule to ban commercial drones or appeal the ruling, the case will allow drone operators to fly their vehicles below 400 feet, at least for now.

And then today: news in that two major US legal practices, LeClairRyan and McKenna Long, have set up Drone case chasing groups. - Washington Post:

LeClairRyan’s drone group, based in Annapolis, is led by Tim Adelman and Doug McQueen, a flight instructor and United Airlines pilot, respectively, in addition to being aviation attorneys. McKenna Long’s practice is headed by Mark Dombroff, a partner in the firm’s McLean office and a former in-house lawyer at the Federal Aviation Administration.

“We want to help [companies] shape rulemaking and get a seat at the table, then actually operate in a world they had a hand in creating,” Dombroff said.

There are thousands of companies building drones and trying to market and sell them, but they are running into hurdles because the federal government has yet to create regulations to govern them, Adelman said.

What a marvellous world......still, as the picture above shows, the hunting season could be prolonged all year

*Update: The FAA have appealed, which means the drones don't fly until its settled, and there will be lots of lawyers droning on about drones

Its not often that two of the fairly unrelated areas we watch come together like this....Google has bought the military Robot maker Boston Dynamics, maker of walking robots (we first featured them in 2008 on Broadstuff) - Engagdget:

The New York Times reports tonight that Google has acquired Boston Dynamics, builder of terrifying walking robots DARPA-related projects like WildCat/Cheetah, Atlas, Petman and Big Dog. Andy Rubin has moved over from leading Android to directing Google's robotics efforts and tweeted a link to the story, commenting that "The future is looking awesome!" While we're sure it does -- if you always thought the T-1000 was just misunderstood -- reactions from meatbags are ranging from slightly uneasy to completely freaking out that a company with robots that go anywhere is teaming up with a company that seems to know everything about us.

On reflection I suspect its Google trying to get a fast track on the sort of technology that will make robot cars more plausible (though why they are doing "civilian" robot cars in the first place is also a matter for some head scratching - you just know the military is going to be the first customer). But as Engagdget notes, it may be for something else far more mundane:

So, which one do you think will arrive first -- Amazon's flying drone delivery service or a Google Now robot that shows up at places it thinks you will be with a backpack full of things you've recently searched for?

Or are we going to be invaded by large walking robotic billboards.....

(Hat Tip Mike Butcher)

As part of our monitoring the Internet of Flying Things, news just in that Amazon is thinking of delivery drones (see BBC video below), and its 1st of December not 1st of April

Amazon, the world's largest online retailer, is testing unmanned drones to deliver goods to customers, Chief Executive Jeff Bezos says. The drones, called Octocopters, could deliver packages weighing up to 2.3kg to customers within 30 minutes of them placing the order, he said.

There are a few inconvenient truths for now, however:

- Most countries aviation regulation authorities are very against licencing civil/domestic non piloted drones, at best its lots of accidents waiting to happen, at worst its a snoopers charter (plus a lot of accidents waiting to happen) - and these are on the big end, one of these through your windscreen would be no joke.
- Drone battery power is still something to be overcome, and they are not that fast, so reach is limited
- It is more cost efficient to load a van with a lot of packages and run a most efficient delivery route algorithm, especially in densely populated areas where drone economics are also likely to be best (time per drop is likely to be the key drone economic measure)

Still, its a sign of the things to come. No doubt early customers will pay a lot for very fast delivery. Longer term, who needs elves and reindeer when you have robots and drones. Or Santa Claus when you have a one-click-to-buy patent and a website.

Broadstuff understands the first Amazon drones will be called Dancer, Prancer, Donner, Blitzen and Rudolf....

I gave 2 talks in 2 recent conferences (its been Conference Season in November, hence blogging dearth) on developments in the Internet of Things world. One was the overall industry, one was based around open source. My overall impression from both conferences is that the Internet of Things is being held up by obfuscation and confusion - every single special interest group/alliance/manufacturer/etc is asserting they are a critical part of any solution (It was ever thus, IPTV was the same in the early 2000's for example).

Of course the market eventually sorts out who is real and who isn't, but it has the effect of making a lot of potential customers sit on their hands and wait until more clarity emerges. This is especially true with networked systems where the cost of device installation or removal is vastly more than the cost of the devices.

So in a nutshell, here is my view of the IoT state-of-play as it now stands - referring to the Broadsight "4 box model" of the IoT above:

Input Systems
There are a very wide variety of devices that will want to connect to this IoT, as noted in the diagram's left hand box. Many are very different to each other. There are few agreed standards for connecting devices of a similar nature, never mind those of a different type. This massively reduces the utility and increases the cost.

Distribution Systems
As above, there is such a plethora of competing technologies that don't Inter-Network easily - as one analyst pointed out to me, the industry is currently better termed "The Siloes of Things" as there is very little Internet happening if many players can avoid it.

Aggregation Systems
Yet again. the aim seems to be to have as much walled gardening going on as possible. Everyone believes in standard platforms, so long as its got them in the centre. Even the open source community is not immune from this, open source does not automatically mean inter-operational end to end. Not all of this is deliberate, some is just necessary - for example, given no standard data taxonomies exist, to make workable systems people just have to get on with it and define their own.

End User Environment
Here too there is confusion, with multiple players asserting their right to be put in front of the end user, though of course many systems "end user" will be bots, decision algorithms and so on. To an extent the difficulty with defining standards upstream is due to the huge variety of end applications, all requiring very different end to end system designs.

So what might the IoT endgame be? In our view the final endgame is a set of common standards, but other industries suggets that there will be several moves in the dance before we get there. Here are some predicted steps:

1. Some siloes will take off early

There are some industry segments where the value that can be created and/or cost saved are too great to ignore, and its worth acting soon. As the only viable end to end solutions today are usually proprietary, these will be used and potentially ripped out (or hybridised) later. Diverse taxonomies will also appear (see the development of the ODI industry into multiple ODI's) at various times for various applications

2. The Open Source industry will take a long time to deliver end to end solutions

As a movement the OS advocates have extremely high creativity and ability. As an industry they they tend to operate disjointedly, building point solutions. The IoT needs highly reliable End to End solutions, and the semi anarchic nature of this industry will mean that this takes a long time - unles a major body forces the speed (cf DARPA with Inter-Networking technologies, or the EU with 2G Mobile standards) and then gives it away for free/mandates a standard

3. Some "poster child" use cases are phantoms, it will take off in unexpected areas

Two poster children are the Automated Home (see our satire on Mrs Fridge), and the Internet of Health. In our view these will be slow to take off as in the former case there is no compelling reason (its been possible for a decade, no one has really bothered) and in the latter, the issue is a combination of data confidentiality and the very severe consequences if systems are not working properly. What we expect is niche use cases - home security rather than home automation, or operating on the margins - in my opinion the "porn case" (dedicated users pushing the boundaries) in this space is coming from the Quantified Self movement, and what they drive will probably be the early Internet of Health applications.

The real early applications will, in our view, be twofold:

- Classic (Geoffrey) Moore "over the Chasm" industries - areas of high returns on any efforts; so the issues of workability, reliability, standards etc are small beer compared to the potential benefits. High value plant and processes or very large reduction is labour costs are obvious areas.

- Easy "Over the top" applications where the application can get onto the Internet early and avoid all the jumble of competing players, and go straight to server-side intelligences. I wrote some years ago about a nursery using old cannibalised mobiles and Twitter to manage sensors, this would be a typical application.

4. A Standard (or two) eventually emerge

Whether it is de Facto (a market dominance emerges, like the DOs operating system or LAMP protocols in the web space) or de Jure (a set of standards is universally adopted) remains to be seen. Looking at all teh parties jostling for position, somehow we suspect it will be de Facto.

5. Everyone will build a Middleware
Where complexity and non standardisation proliferates, then the temptation is to build a middle ware layer that just translates all the functions of disparate systems and presents them as a common UI. Only problem is, everyone wants to own the middleware layer so there will be multiple non-collaborational middlewars

6. Privacy/Hackability is going to be a major issue
There are two things that will drive this - firstly, many of the projecst mooted seem to be about building things because it is possible, rather than because it's what users/citizens/employees etc want, and these systems can impinge on privacy. Also, many of the early devices sare being built with a sort of naive assumption that just because no oane has hacked them yet (abeacuse they don't exist/there is no value etc) no one will. As any securty system expert will tell you, increasing the Internet by 50 billion odd devices increases the security risk by the same amount

In essence I think we are in for a few years of smoke, mirrors and obfuscation and if I had to bet on a first "Proper Internet" of things. it will be largely "over the top" and using as few systems as possible.

TED talk demonstrating the effectiveness of complex algorithms on a small quadcopter's flight capability, and how it can be used collaboratively with other quadcopters and human systems to the point where advanced technology starts to look like magic.

Things you need to take away are:

(i) The processsing power is all onboard
(ii) Very powerful algorithms are the weight of the processing power
(iii) The Algorithm trupms the aeroplane. Software >> Hardware...
(iv)...once the Hardware is up to the job. This sort of onboard battery and power power wasn't around when this blog strted in 2006
(v) Hand commands for robotic devices are here, eye commands will be next.