2010-06-10

Spirit Sol 511

"My Ph.D. thesis was on writing a program that would design robots for you," Chris says. "You'd tell it what you want, and it would design a robot to do it. That way, I could spend more time climbing." (He's an avid rock climber.)

"See, I'd have done it the other way around. I'd have designed a robot that would go climbing for me, so I could spend my time designing robots," I say.

"That's the difference between you and me," he deadpans.

It's Friday, and that means it's another three-sol plan. I did essentially all of the IDD work for all three sols yesterday so that I could focus on the drive today, which should make today easy.

Ha.

Since we're on tight sols, our uplink's at 17:40, and for a while there it really doesn't look like we're going to make it. In the end we finish with less than half an hour to spare. We haven't cut things that close in a while. I am awfully glad I did the IDD sequences up front, though; it was only two or three hours' work, but we really didn't have those two or three hours today.

This might be my last sol on Spirit, at least for the next month or so. I'm potentially on jury duty next week, and then I head upstairs to the Land of Opportunity. Which will be different from the last time I was up there. I came downstairs not long before they ran into what they're now calling "Purgatory Ripple," and they extracted themselves already. But the effects of Purgatory Ripple will outlast the actual experience of our being embedded in it. We're going to spend some time investigating the ripple itself, and we're going to drive differently from now on -- probably much less aggressively, but certainly differently.

How differently? That's still an open question. Khaled's working out some of the details, and he comes by to talk to me and Chris about it. They've got a list of proposals going, including my two sneaky tricks. (First trick: periodically stop and try to turn in place. If you can't turn, you're stuck; preclude further driving. Second trick: periodically try to run visodom. If it converges, you must not have moved significantly between the two images, meaning that you're stuck; preclude further driving.) As we talk, I come up with a couple of other creative ideas. One of them is to abuse the IMU: if it doesn't show that your local "down vector" is changing with respect to time, you're probably not climbing ripples and hence may be stuck; preclude driving. Similarly, we could check the suspension limits: if the suspension isn't flexing over time, stop driving.[1]

Nearly all of our ideas are reactive measures: try to detect when you're bogging down and keep from making a bad situation worse. What we'd really like are predictive measures, where we avoid getting stuck in the first place. We're short on those.

But we'd better do at least some driving. Recent email traffic prompted by the 1000-sol news has pointed out that we're close to a couple more milestones: 95,303 images so far (nearly 100,000) and a total distance of 9752.26m driven so far (nearly 10,000). I'm so proud of these rovers.




Footnotes:

[1] The end result was to use VO in a cleverer way: take a picture, move 20cm, and update VO. If VO reports that you've slipped more than 40%, preclude further driving. If it doesn't converge, try again, and preclude driving if you slip more than 40% or if it still doesn't converge.

But that's expensive: it takes at least six minutes, as many as 10, and we have to stop and do it every 20m or so on Opportunity. So, as we speak, there's an MIT grad student working to make an IMU-based classifier do the work for us -- an essentially free check that will notice relatively quickly if we're embedded and stop driving.

[Next post: sol 523 (Opportunity sol 503), June 23. Damn jury service.]

No comments: