Robotics

Who could resist? From NASA.....beats programming throwaway sheep, Tim O'Reilly would be proud

We are looking for a full-time developer to help NASA send robots to the Moon.
If you want to make a difference in space exploration, come join us!

Ideal candidates will have a solid history of implementing and deploying
complex software systems that integrate multiple languages and libraries on
multiple platforms.

Responsibilities:

- Develop software tools for remotely controlling a rolling-walking robot.
This project includes interactive 3D graphics, GUI design, middleware, and
working with multi-language/multi-platform components.

- Develop software sysetms for processing satellite imagery of the Moon
into maps and 3D models. This work involves creating very high-performance
processing pipelines and innovative geospatial user interfaces.

Dusting off all those notes from my BSc final year Robotics stuff ......

(link hat tip Ariel Waldmann)

Readers of this blog will know we follow developments in robotics. Here is one such (from the Daily Mail, via Nick Carr).



AI/IT pioneer Alan Turing devised the Turing Test to define when we may know a device is intelligent - when it can fool a human into thinking it is one. His test was pre internet, so needed Sneakernet (or in his time, Shoenet). The test says that:

....a human judge engages in a natural language conversation with one human and one machine, each of which try to appear human. All participants are placed in isolated locations. If the judge cannot reliably tell the machine from the human, the machine is said to have passed the test. In order to test the machine's intelligence rather than its ability to render words into audio, the conversation is limited to a text-only channel such as a computer keyboard and screen

No one could tell you were a dog on the text internet, it is true. But a project, called 'Human-Robot Interaction', was devised at the Bristol Robotics Laboratory (BRL), run by the University of the West of England and the University of Bristol. A team of robotics engineers - Chris Melhuish, Neill Campbell and Peter Jaeckel - spent three-and-a-half years developing the breakthrough software to create interaction between humans and artificial intelligence.

So look at the above video, and make your guess about how long it will take before machines can pass the Turing test on a video channel. 5 years? 10 years?

I was reading Chris Anderson's article on the size of the FreeConomics market (of which another post later) when I came across his other blog - for amateur Unmanned Aerial Vehicles (UAV, or Drones) design and lost a happy hour there. Chris is thinking of setting up competitions for amateur UAV builders.

UAV's combine the Geek Triathlon of mechatronics/robotics, complex computer systems (for autopilots etc) and aeronautics (there is a parallel strain with sailing vessels) coupled with model building - who could resist?

There is some great stuff on the site about converting ordinary shop bought models - planes, choppers, blimps - to UAV usage. But this small, flapping UAV - with camera inserted - from the University of Delft takes the prize.



(Amateur UAVs have already crossed the Atlantic - see here)

(Disclosure - at Uni yours truly initially studied Aero Eng and has designed and even flown - without immediately crashing - a number of R/C model aircraft, and final year BSc design project was a 6 axis robot - I was only seduced by the Internetz for an MSc. Hence the odd article on mobile robotics on this blog

Fascinating video of the DARPA/Boston Dyamics legged robot walking independently.....



I'm interested mainly because it has 4 legs, not 6. The initial thinking behind "proper" walking robots was that they would be 6 or 8 legged, since insects / arachnids require less computing power to use 6 / 8 legs than higher order animals need for 4. Also, 4 legs has no limb redundancy. For those who follow the development of robotics, here is a simple summary of leg from MIT's work on planetary surface explorers - the "Little Martian Robots" (LMR's)

2 Legged LMR
Many bipedal robots have been designed. For a selection of some of these intriguing machines visit the MIT Leg Lab Website. It is conceivable that these robots could be designed to be stable without the need to stay in motion. The martian surface, however, will not be particularly flat making bipedal balance a fairly tricky issue. Also, the Martian winds and weather would make it difficut for a bidepal robot to stay standing. Should one of the legs break, the robot would become an imbalanced mono-legged rover. Such a robot would be incapable of balancing whether in motion or in stand-by mode.

4 Legged LMR
4 Legged LMR Quad-legged robots like the Leg Lab's quadruped, have more realistic leg configurations. It is fairly easy to balance a four legged robot, even on a fairly irregular surface. The rover would be capable of motion across the Martian landscape and would have the ability to stop and remain standing. A four legged LMR, however, would not be very rugged. Should one leg break, the robot would find itself imbalanced and fairly incapable of moving. Indeed, a three legged rover would need to be programmed in an entirely different way than a four legged rover, making it necessary to include a separate control program in a case a leg were to break off.

6 Legged LMR
6 legged LMRs like the MOBOT Lab's Hannibal and Atilla are more realistic models of legged motion. The success of the Hannibal and Atilla prototypes clearly show that a 6 legged LMR would be capable of functioning on the Martian surface. It would be capable of balancing well, even with strong winds and the Martian dust storms. The LMR would also be capable of sustaining itself even if one or (under certain circumstances) two legs were broken. There would be no need to completely reprogram the LMR as the models of motion for a six legged LMR and a five legged LMR are very similar.

8 Legged LMR
8 legged LMrs would allow for the greatest usable level of redundancy on the LMRs. It is conceivable that the LMR could continue functioning if up to four legs are broken. This should give the small LMRs the capability to stay out in the field for longer periods of time. A more important use of the extra two legs of an 8 legged-LMR exists however. It would be possible to use any pair of legs as a proximity detector (Angle and Brooks, 1990 "Small Planetary Rovers" ONLINE). If the leg were to brush against anything, the LMR woulld be capable of noting that it is close to an external object. In this way, the safety of the LMR in-motion can be gauranteed with more certainty. It could also be possible to carry sensors of some sort on the non-critical front two legs. Indeed a whole variety of options are opened up in having an 8 legged LMR.

We have chosen that our small LMRs use the 8 legged system over a 6 legged system for the above reason.

A remembrance in the week of Arthur C Clarke's death - I was one of those kids of my generation fascinated by (i) Space Travel and (ii) Robotics, and thus especially space travelling robots. However in the 1980's when I did my BSc and MSc Engineering degrees, the computing power and various associated tech. (eg battery power) required to make these things work just was not there (My final year design project was a robot btw) - it was sadly clear that we had to wait quite awhile, so I turned my interest to the emerging networked computing field. In the 1990's we also saw the emergence of Evolutionary algorithms, Neural networks, fuzzy logic etc which have really pushed the capability of self-controlling automata

(Thanks to LLoyd Davis for link)

We are doing a piece of work right now where the evolution of RFID and M2M (machine to machine) services is a part of it. While working on the mobile aspects of this, it occurred to us that most people conceive of the m2m device / agent / whatever as quite dumb - a pallet, a piece of inventory, a truck etc etc, and its main role is to tell something more intelligent about itself.

But what happens if the device itself is (semi) intelligent, and is using the m2m comms as much for its own purposes as much as anything else. One of the interesting insights from work on Artificial Life is that often the combinations of fairly simple rules allow very complex behaviour, and quite a high level of self organisation.

One fairly common example is where the common or garden house vacuum cleaner robot may use m2m comms to tie up with other devices to define when and where it cleans. But as Moore's law progresses these devices can all have more intelligence, and this can allow them to exhibit fairly complex capabilities even for simple devices.

For a start, they could use evolutionary algorithms to learn new tasks - like learning to fly for example:

Learning how to fly took nature millions of years of trial and error - but a winged robot has cracked it in only a few hours, using the same evolutionary principles.

Krister Wolff and Peter Nordin of Chalmers University of Technology in Gothenburg, Sweden, built a winged robot and set about testing whether it could learn to fly by itself, without any pre-programmed data on what flapping is or how to do it.

To begin with, the robot just twitched and jerked erratically. But, gradually, it made movements that gained height. At first, it cheated - simply standing on its wing tips was one early short cut.

After three hours, however, the robot abandoned such methods in favour of a more effective flapping technique, where it rotated its wings through 90 degrees and raised them before twisting them back to the horizontal and pushing down.

"This tells us that this kind of evolution is capable of coming up with flying motion," says Peter Bentley, who works on evolutionary computing at University College London. But while the robot had worked out how best to produce lift, it was not about to take off.

"There's only so much that evolution can do," Bentley says. "This thing is never going to fly because the motors will never have the strength to do it," he says.

.

(report from New Scientist)

That was 2002...today, there are robots that can fly like insects - these tiny robot flies (see pic below) for example. Imagine swarms of these operating using self organising behaviour routines (flocking, direction finding etc), each built with a different type of very simple sensor.



But thats just for starters...how about this one - unmanned Aircraft with Solar cells that can in theory cruise the stratosphere in large numbers, using self organising software to keep their coverage going at high levels of reliability - who needs satellites for TV signals, mobile signals, ground surveillance with swarms of these.....the economics vs satellite (or even mobile base station networks) might be pretty interesting.

(The pic below takes it even further - using these to explore Mars)



Now...mash some ideas together - RoboFly RFIDs, Glider Grids....bears thinking about, doesn't it?

Yes, it is "way out there", but it is (theoretically) doable, and it is Friday afternoon after all.