[X] My Assistant

[Stu]

A lot more detail about the "Maxwell Plan" here...

http://roadtoendeavour.wordpress.com/2010/...al-to-the-metal

[PaulM]

...two 70m drives back to back, which is what most of us and the rover team wants.

I understand that if wheel damage was not an issue then it would be easy to plan 140m drives for Oppy. This has restricted Oppy's progress to 70m per day.

What has occured to me is that it would be possible to plan two rather than one 70m drives at a time and on the second day to simply to perform the second 70m drive without waiting to analyse end of drive photos. The second day's drive would be completely safe because the previous days photographic coverage from no more than 140m away would be adequate.

My plan would largely eliminate the problem of restricted SOLs.

What I have observed is that often a 70m drive day alternates with a resting day. If my plan was adopted then a cycle of driving for two days driving followed by one day resting would result in increasing the ground covered every six days from 210m to 280m.

[jamescanvin]

The second day's drive would be completely safe because the previous days photographic coverage from no more than 140m away would be adequate.

No, the limit of 70m is because that is the maximum distance they are comfortable doing a blind drive in this terrain, to do any further auto-nav is needed which up to now has required forward drives. It is these forward drives which are a problem for the wheel currents. This is why Scott is proposing this complex plan of backwards-autonav to go further without the rover drivers getting the images and checking that it's safe. When we are out into the 'parking-lot' terrain then it is possible they could push the blind drives a little further, but I doubt they'll be >100m

[ElkGroveDan]

Since we are talking about blind drives here one thing worth mentioning is that when the concept of blind driving was first planned for, the best ground resolution they had was the MGS MOC with an optimum resolution of 1.5 x 1.5 m/pixel. So the chance of surprise encounter with problematic ground relief was significant. After landing we were then fortunate to acquire some nice MOC cPROTO derived images which had a resolution of 0.5 x 1.5 m/pixel where you could start to see some smaller ground features as in this shot of Spirit at Bonneville Crater. Now we have the HIRISE images with a resolution of something like 0.25 m/pixel which allows us to see the rover tracks quite clearly and in hindsight we can identify objects like Block Island from orbit.

So one of the things I've been wondering is if they have updated or narrowed the original safe driving parameters or the rules for blind drive segments in light of the ability to have a decent look at the planned drive route from above.

Secondly has there been any though of applying super-resolution image stacking techniques to a series of HIRISE images to further improve on the ground resolution which could assist in safely loosening some of the blind drive parameters. I'm sure this would require some immense logistical planning and targeting to acquire enough images under identical lighting conditions. Has anyone heard if the concept of stacking multiple HIRISE images for improved resolution has even been discussed?

[climber]

I'm pretty sure this issue has been rosen here (before MRO launch IIRC) and the answer was that the technic used with MGS was not applicable to MRO... but I can't remember why...and I think it was an answer from Doug. I may be wrong but...

[stevesliva]

I'm pretty sure this issue has been rosen here (before MRO launch IIRC) and the answer was that the technic used with MGS was not applicable to MRO.

Can't do cProto-- I think it involved slewing the camera or spacecraft-- with a pushbroom camera, but you can probably stack.

My idea with hirise was to take late day images with long shadows to emphasize drift topography, but even that's a tall order since it passes overhead at the same time every day.

[Stu]

Apparently my blog has been playing up today, so if anyone has tried to access the post on Scott's plan but been unable to, here's what he wrote me:

"You probably know already that we used to do ~ 70m blind drives,
followed by an autonav segment. (Autonomous hazard avoidance, that
is.) We had to quit that because of the RF wheel currents: HAZCAMs
don't give good enough range data in Meridiani terrain to support
autonav, and the NAVCAMs have to look *forward* for autonav, because
behind us the LGA is in the way -- but we can't drive forward any
more, because it makes the RF wheel currents rise. But I thought of a
possible way around that limitation.

"Roughly, if the experiment works, here's the eventual procedure.
First, we do a long blind drive (just a little shy of 70m), followed
by a slip check. Now we're good to go another 20m, the normal
distance between slip checks.

"We slew the NCAMs around so that they're almost seeing the LGA, but
not quite, an azimuth of 162.5 degrees in rover-frame. Now we turn
*Opportunity* the remaining 17.5 degrees (clockwise) so that the NCAMs
are looking straight along our intended drive path. We image the
terrain for hazard avoidance and turn back 17.5 degrees anticlockwise.

"At this point, we're aimed (that is, our rear end is aimed) at our
destination and we have imagery in the drive direction. We do a
straight-backward step of 1m (actually 2 x 50cm steps, for technical
reasons) toward our goal, telling Opportunity to take the step only if
the hazard-avoidance imagery we took shows it to be safe. Then we
just repeat this turn/image/turn/step loop for 20m worth of driving.

"It's a little more complicated than this, because we first have to
drive about 2.5-3m blind to get onto the nav-map data (the nearest
stuff we can see looking over the rear solar panels is about 2m away)
and for other reasons, but that's the basic gist of the idea. If it
works, we can go 90m/sol rather than 70m/sol, a 29% drive-rate
improvement. There will be sols where this doesn't work -- we're
heavily constraining what Opportunity is allowed to do, and if she
sees something scary, this procedure doesn't let her drive around it;
also, the turns-in-place might elevate the RF currents, in which case
we'll have to abandon it; and there are other things that can go
wrong. But if this worked perfectly every time, it would cut nearly
40 drives out of our trek to Endeavour -- a 2-3-month savings, at
least, probably actually more than that.

"In a real blue-sky future, we can do a slip check at the 90m point,
followed by *another* 20m of hazard-avoidance driving, and so on. We
can have hazav-only drive sols, similar to what we did over
President's Day weekend in 2005. When the terrain is a little
friendlier, we can do longer blind drives before the hazard-avoidance
segment(s). And there are other possible optimizations. Right now, I
want to perform a basic experiment to see if this even works."

A HUGE thanks to Scott for taking the time to write and explain all this to me, and for us.

[ElkGroveDan]

I wasn't referring to the MOC cPROTO "pitch and roll" orbiter technique which is entirely not applicable to HIRISE. I was referring to super-res stacking.of multiple images taken the same time of day.

[nprev]

Thanks for sharing this, Stu, and of course many, many thanks to Scott for taking the time & effort to inform us so thoroughly & clearly!

The super-stacking question is interesting, Dan. How far can they go? Based on the discussion between Ralph & Jason over on the Titan thread (EDIT: which for some reason I can't find right now), it seems that 'resolution' is a fairly complex concept in many ways.

[djellison]

I don't know is super-res techniques can be applied to bushbroom cameras at all. Framing cameras, sure, but pushbroom - I don't know. You'd have to map-project before you stacked, so at that point I think you're already working with 'pixels' that never existed anyway.

[mhoward]

Good news: Exploratorium may be back up and running. Trying it out now; fingers crossed.

[Stu]

Just had an email from the Exploratorium webmaster telling me that things should be up and running over there again :-) Probably a whole load of files in one monster folder, but good to be catching-up...

[fredk]

A whole bunch of old images and two new ones (that I noticed) are through. The new ones are from the 2301 drive:
http://qt.exploratorium.edu/mars/opportuni...0M1.JPG?sol2301

[Phil Stooke]

"You'd have to map-project before you stacked, so at that point I think you're already working with 'pixels' that never existed anyway. "

Doug is right about this, but the situation is not hopeless. I would work just with a small area, not a gigantic full frame. OK, so I have several HiRISE images of the same area, preferably a flat area so relief distortion will not be an issue. I would enlarge each image before doing any reprojection, so I'm not losing any information by reprojecting. When they are all perfectly registered I would sharpen them a bit and combine. The enlargement is done for super-resolution anyway, by factors up to 10x, so the only difference is the geometric reprojection, and as I say if you reproject after enlargement no information is lost. Tim Parker has a good description of the method somewhere, should be easy to find. I used it with Voyager images of, among other things, Hyperion.

You can do some interesting experiments with this technique by taking one high res image and subsampling slightly different crops of it - copy a 200x200 selection from several locations shifted by just 1 pixel each. Copy each one to its own file, and downsample it to 100x100 pixels. Now you have several different images all copied from the same source but from slightly different locations. Enlarge each one back to 200x200 using nearest neighbour interpolation and look at small details - they are sampled quite differently. But using super-resolution recombination you can get quite close to the original image.

Anyway, no reason why it would not work with HiRISE. Maybe I'll have a go.

Phil

[ElkGroveDan]

Maybe I'll have a go.

You might select an area where we already have ground truth imagery and a wide range of < 1.0 meter objects to inspect the lower boundaries of the improved resolution. Possibly the cobbles around the rim of Victoria.

I'll pay your Starbucks bill for that day.