Smart Satellite Sets Its Own Priorities

Roland Piquepaille writes "Currently, satellites take pictures of whatever is in front of their cameras. But hydrologists from the University of Arizona (UA), working with the Jet Propulsion Laboratory (JPL) are creating spacecraft that think for themselves. Their smart software, which is tested on NASA's EO-1 satellite, can be used on all kinds of spacecraft. This software has three components: an image formation module, a science algorithm module, and a continuous planning module. This onboard planner reschedules what to film in conjunction with what the scientific algorithms have detected. This software has already detected floods in Australia and will be adapted to also detect volcano eruptions and changes in ice fields. More details and references are available in this overview, including images of the flood detected by this smart software."

Comments

Filtering software

[PineHall]

Isn't this what software here on earth can do and are doing? Putting it on the satellite does not change anything. I think you would want the satellite to send all the data it collects, so why not filter it here on earth. If the satellite sends only the data it finds interesting, it will miss some events that it was not programmed for but would be useful to the scientific community. Send all the data and filter it here.

Re:Filtering software

[StarWynd]

I agree that having filtering software on the spacecraft rather than on the ground does not change anything for this particular project, but there are some situations where having such software would be very beneficial.

If your spacecraft has a limited bandwidth where you are forced to throw some data away, you will want some onboard processing to determine what's "interesting" or not so that you will have a better probability of getting better science data on the ground. Such software is also vital for spacecraft which have capabilities similiar to Deep Space 1's autonav system. Imagine the possibilities of a spacecraft that can fly itself with very little ground interaction and able to automatically determine which instruments to use and when. Currently, deep space missions get planned out years and years in advance in order for the spacecraft to be completely utilized all the time. It would be nice to at least have an "autopilot" feature for not only attitude control but also automatically find opportunties when the instruments could be best utilized.

If you have a typical earth orbiting satellite with a high transfer rate, just return all of the data and do the processing on the ground. We have a number of large databases of satellite data just so we can do our own filtering and analysis on the data years and years after the fact. Some of our processing requires days or weeks to execute and sometimes we still don't know if the data is "interesting" or not. If we let the spacecraft determine everything, there will be things we miss. I'd prefer to use the flight software to only gather data and package it up rather than try to make fancy decisions for us. Of course, all of this also depends on the role of your data. Certain datasets will benefit from such advances. My data won't -- we need as much of it as we can get. But if we didn't have a large data rate, the best possible solution would then be to put some processing on the spacecraft to increase our odds on getting good data.

What does all this mean? It means that you need to do what's appropriate for your data and as always, your mileage may vary.

Reminds me of "robot scientist"

[FleaPlus]

This reminds me a lot of the robot scientist from earlier this year, which was able to formulate hypotheses and perform experiments to determine the metabolic pathways of yeast. I'm quite excited about where this sort of technology can take us in the future, removing much of the drudgery that grad students/technicians have to do and accelerating the advance of scientific progress.

Gaze control

[Animats]

Gaze control is important, but far more useful in earth-bound systems. A good application would be to use it with surveillance cameras and traffic monitoring cameras, so that the interesting stuff is presented to humans, while endless pictures of empty rooms and smoothly flowing traffic are ignored.

Sounds like a TIVO but

[Zukix]

So it can learn from what sort of things you have asked to observe in the past and have future unrequested data flagged as interesting? Very difficult for it to actually produce useful results. Ground based crunching of a vast data glut from a large constellation of inexpensive dumb sats with lots of redundancy would seem more appropriate with ground based commanding and intra constellation communication to handle sats that are out of contact (interesting orbits are not geosynch)

Its an interesting challenge to be responsive to variable priority planning requests from multiple clients some of whom can request 'in theatre' with mobile transmitters not just at permanent ground-stations. Really hard computational problem with strict time constraints and lots of factors such as power-up times, manauover times etc.

I had a nice idea of auctioning satellite time so that you have to pay more to bump requests. Disruption to a schedule by a new request would be factored into the cost of accepting the request but the various clients would be put into financial competion to outbid for service. Use the market to schedule.

Re:What's the big deal?

[georgewilliamherbert]

Other autonomous duties don't seem like such a stretch when it only takes a second to communicate

They are a big deal.

Spacecraft control automation has been a huge problem for decades. The ability to manage failures and continue degraded operations rather than safemode the spacecraft (and stop collecting data in many cases) is still unproven.

Not surprising.

[NegativeK]

It was really only a matter of time before automated image selection moved to further applications. From what I understand, Fermilab has been doing a very similar thing - with millions of "images" from each collision, the _only_ way to look at the remotely interesting ones is to have an automatic selection process.

It does, however, make you wonder about the really interesting things that could be missed in the process.