Hello again! :-) And an Update on my UAV Project. Any Thoughts Welcome...

[FirmamentFX]

Hi guys,

It's been a long time since I was on here - unfortunately life got in the way and threw my UAV project off balance for a while...

Anyway, I have been thinking about a fully automated integrated autopilot/takeoff/landing system for my UAV over the past couple of months.

I have decided to completely "over specify" the computer systems in my vehicle to allow for future growth and to allow me to try out some more complicated ideas I have planned.

To that end, there will be 3 SBCs in the UAV, all running Pentium M processors with 512MB RAM (I got a great deal on some second hand boards and have found some nice cheap processors...). They will also all be running Windows CE (the hard, realtime version of Windows). The breakdown of processing is detailed later.

On the whole Windows thing (I remember some discussions from the latter half of last year regarding RTOSs) - I looked at several options including QNX, VxWorks, TenaSys RTX for XP Embedded, and Ardence RTOS. I eventually came down in favour of CE because it includes pretty much everything one would need, whereas the other systems charge for "bolt ons" such as networking capabilities and advanced graphics (necessary for blob tracking for machine vision and real time video streaming).

I can also develop in Visual Studio 2005 in C and C++, which I am familiar with - this was another deciding factor in the choice of OS. A lot of code will be generated using MatLAB and various extensions, before being brought into VS2005 for integration with other control code and compilation.

The breakdown of the processing functions is as follows:

System 1: The flight and system controller, which deals with flight control, flight plan management, flight augmentation, and all the sub systems including fuel/pneumatics/electrics/gear and brakes/etc.). There is also a secondary GPS system attached to this computer for return to base and redundancy (in case System 3 goes down).

System 2: Stores Management System, which deals with the control and processing of data from the payloads (for example cameras and other externally looking sensors) and also the control of the payload bay doors etc.

System 3: Data Acquisition Concentrator, which takes all the data from all the onboard sensors (positional, air data, Primary GPS, radar altimeters, INS system, and the "pilot" camera) and manages the encoding/decoding of messages from the ground control ststion (the radio modem is connected to this system). It also manages data logging into a database in the system.

All these systems are networked together using an onboard LAN (I looked at CAN bus and other data protocols including RS232, but the data flow will be very high).

Also networked into the system is the guiodance camera, which manages the blob tracking for the landing system. This is a self contained 3rd party unit that communicates by ethernet and also runs Windows CE with custom programming).

As I said, I know this is perhaps over-specifying the processing power needed, but I thought I may as well to allow for future upgrades as I got such a sweet deal on the electronics... :)

Wow! Long post. Sorry :) . However, it's great to be back working on this project, after a rough and bumpy few months in my personal life! :rolleyes:

Cheers!

Martin

[Tom Harper]

Sounds great!

What's going to carry all that stuff?

Tom

[FirmamentFX]

Initially I will be using a large scale commercially available aerodynamically stable RC aircraft, but I eventually want to put it into a custom built model.

Something else I am *very* interested in is the idea of a distributed AI neural network (basically a swarm of UAVs). This is something I am developing as a simulator model (using Simulink) and will probably be developing for several more years before it becomes a reality! I would like to apply it initially to a 3-UAV setup (lead and 2 wingmen). This is why I want massive onboard processing power...

Cheers,

Martin

[dalbert02]

What is the UAV mission? I am not sure if you are posting and looking for suggestions or just offering a progress report. Regardless, please keep us updated, it certainly sounds interesting!
-dave

[Tom Harper]

Sounds very ambitious. I've considered neural nets for automatic stabilization - Op amps with resistors for weights. A few of diodes and resistors will do any sigmoidal function. The trouble is that the problem is easier to solve with a cheap micro.

[FirmamentFX]

Hi Dave,

There is no single "mission" per se... In very general terms photography and essentially as a test bed for electronic systems.

The post was in the main a status report having been "off the board" for several months! But as always suggestions are welcome!

Tom, I am thinking of neural nets in terms of software controlling the onboard flight control system. There is some pretty cool software that will spit out the C or C++ code for a neural network you design...

A long term idea is to look at the use of distributed neural nets over 3 or more aircraft to see if they can "intelligently" react to situations like flying in formation or "attacking" (not *really* attacking :) ) a target by operating with each other using standard military strategies and manoevures.

AI is something I am becoming more and more interested in (to be honest Microsoft Robotics Studio got me into it - a robotics mad friend of mine pointed me to it), but first I need a stable flight platform to develop these ideas on, so the AI side of things is on the back burner for the moment - still being thought about though!

As I mentioned a few months ago, part of the development is a high-level mission programming language (where you can simply describe procedures like "Land", "Takeoff", "Circuit" etc and the AI manages and adjusts for the variables for you).

M

[hg1]

I've considered neural nets for automatic stabilization - Op amps with resistors for weights. A few of diodes and resistors will do any sigmoidal function. The trouble is that the problem is easier to solve with a cheap micro.

It would be a lot easier to code the neural net in software. In any case, this is a worthwhile approach to consider. I've seen several technical papers which compare the performance of neural nets favorably with extended kalman filters, and the former doesn't require any modeling of the system.

[FirmamentFX]

Neural nets would be great for auto stabilisation - especially for V/TOL aircraft.

Not sure if I'm going to do the F-22 or the F-35 for my "custom" model at the end of this though :) (bear in mind, this is a project for fun too, so the fun part for me is building the UAV as a *real* scale aircraft - I thought about doing an airliner, but a fast jet is a hell of a lot more interesting, and challenging). The idea of a V/TOL UAV is fab, but the F-35 is such a darn ugly piece of kit... And the harrier is far too complex.

M :)