ExoMars Options

[Guest_Sunspot_*]

http://news.bbc.co.uk/1/hi/sci/tech/4180840.stm

Europe has fixed on a concept for its next mission to land on the Red Planet.

It aims to send a single robot rover to the Martian surface along with another, stationary, science package.

[djellison]

Of course, with custom realtime OS's - the processing overheads for your average spacecraft are only a fraction of those for the OS's used by those 'mainstream' processors. I've not actually heard of computing performance being a limiting factor for spacecraft - but I may have missed such reports.

Doug

[mcaplinger]

I've not actually heard of computing performance being a limiting factor for spacecraft - but I may have missed such reports.

Generally, Doug is right. There's a lot of semi-informed speculation on this thread, less real info. The RAD750's performance is comparatively poor from two factors: first, the process changes that make its internal registers immune from radiation-induced bit flips slow down the clock speed considerably, but more importantly, external components, also rad-hard, are running more slowly, as are the busses. The RAD750 on MRO doesn't even have an L2 cache and it's using a 33-MHz PCI bus.

If you wanted a non-mission-critical computing resource that didn't have to be totally bulletproof against radiation, there are many options, including commercial processors that happen to be latchup-immune and various gate arrays. For our MSL instruments we are using Xilinx FPGAs; clocked at 40 MHz they are many times faster at doing JPEG compression than code running on a fast desktop system would be.

Rover speed is typically limited more by the capabilities of the drivetrain and the overall power budget. It's not like MER would be going 50 KPH with a faster processor. Despite what AI people will try to tell you, we don't know how to write autonomous nav software regardless of how fast our processors are.

And finally, MIPS (aka "Meaningless Indicator of Processor Speed") is a bad metric for judging computer performance in this or any other problem domain.

[RNeuhaus]

Rover speed is typically limited more by the capabilities of the drivetrain and the overall power budget. It's not like MER would be going 50 KPH with a faster processor. Despite what AI people will try to tell you, we don't know how to write autonomous nav software regardless of how fast our processors are.

And finally, MIPS (aka "Meaningless Indicator of Processor Speed") is a bad metric for judging computer performance in this or any other problem domain.

I was thinking that too. The AI is one of the software components which needs a fast CPU, lots of RAM in order to perform the harzard avoidance analyze more sophisticated and perform the required action with a much better performance as the MER does. A much improved AI will need much less from Earth remote direction and hence the rover will have greater autonomy to perform the core activities more productively in Mars.

It is true that the MIPS "Millions Instructions per Second" is an old comparision computing power that actualy is obsolete except it is only good to have an idea about how the younger brother computer is improved against the older brother if the model or serie is about the same.

Well, I seems like that the AI is a new field that must work harder to improve the space exploration by improving the autonomy capabiltity of probe or rover. If the microprocessor RAD750 is limited in its computing capability, so why don't put more microprocessors in parallel. The most powerfull computers work with many processors in parallel.

In few words, I think the AI is still very new and I speculate that in the future, the AI will play with a much greater importance. Imaginate that JPL tell the rover: "Please go there, over that dark spot and tell me what is that up?

Rodolfo

[mcaplinger]

If the microprocessor RAD750 is limited in its computing capability, so why don't put more microprocessors in parallel.

Three answers: mass, power, and cost. A single flight RAD750 board uses tens of watts, weighs over a kilo (just for the board, not counting card cage, etc.) and costs, last time I checked, nearly a million dollars. And we don't need more cycles anyway.

These days, I would take a look at AI development in the computer-game industry.

Games have driven graphics development, sure. But I would argue that there's nothing like real AI in any game out there. Real AI of a sort useful for rovers would be able to sense the environment and react/plan accordingly. Games just don't have to do that; they define the environment, there's no need to sense it.

[RNeuhaus]

Three answers: mass, power, and cost. A single flight RAD750 board uses tens of watts, weighs over a kilo (just for the board, not counting card cage, etc.) and costs, last time I checked, nearly a million dollars. And we don't need more cycles anyway.

Good to hear your comments!

About the weigth, its is by far heavier than any normal microprocessor and its peripheral components; its price is prohibitive for any commercial applications. However, a more powerfull microprocessor will save money on the other side. It is that we are going to learn the results quicker and hence the mission won't take as long as does MER, hence saves money to the mission operations.

Hence, I see that AI is a very promisory role for future space missions and NASA must pay greater efforts on that. I have enclosed a interesting reports in which make lots of emphasis about the importance of autonomy for a greater producivity of mission. The productivity depends much by a powerfull microprocessor.

For rovers and robots, we're trying to design autonomous intelligent agents that can survive in hostile environments.

Mars is a lot more complicated than that, but this particular technique is based on trial and error, so it's self-learning. We train the robots with something called a "fitness function," but we're not to clear on how to build to most optimal training regime. We want to mix and match different types of environments to get the robot to learn to be robust, so no matter what situation it finds itself in, it can still navigate.

But spacecraft engineers are notoriously conservative, and they don't like new things. So it's a constant battle to try and convince the agency that what we're doing will work and that it's better than the technique they currently have. That's always an uphill struggle.

So AI is still a novel and with a radiation hardened microprocessor up to date will help to improve the AI.

A Naturally Inspired Robot MarExo

The problem resides of a very long time lagging between the new microprocessors and the radiation-hardened ones. Mike, do you know why it is?

Rodolfo