October 15, 2004
Guanaco Camp, Atacama Desert, Chile

Agenda
• Conduct autonomous navigation experiments
• Test rover executive fault recovery

Status and Progress
Conducted autonomous navigation experiments. Today we conducted as series of autonomous traverses across a plano cut through by arroyo (dry streambeds). These streams flow only with rain or snow melt that occurs on a decadal cycle but they represent an interesting obstacle to navigation. There are two challenges that Zoë’s faces: detecting these continuous, negative (hole rather than hill) obstacles from a distance and negotiating the descent and ascent from the streambed when it must be crossed. In the experiments today (about 33) Zoë was conservative, often wandering around to find an ascent point when charging up the slope would have succeeded, and we have found some weaknesses in the current navigation technique when confronted with this type of terrain. Most of the (33) experiments today ended either when Zoë decided to make a sharp turn on a slope or when we decided to make an modification the action execution algorithm to improve performance. We’re compiling statistics on the cause for the termination of each traverse (sometimes it’s because Zoë reaches its goal!).

Exercised recovery behavior. The real excitement of today was the performance of the navigation recovery behavior. During the 2003 field season, we prototyped a behavior (meaning an action that is not planned in advance but one that is initiated in reaction to current events) to enable the rover to back up for another look when all paths ahead appear blocked. This year that algorithm has been refined and today, really put to the test. Zoë is surprisingly persistent in backing up, usually just a few meters but sometimes repeatedly, and trying a new path ahead. Having all paths blocked, which can occur with a few strategically placed rocks or when encountering the stream embankment, was a significant cause for traverse termination last year. Part of our success in continuing to increase the average and maximum distance of single-command autonomous traverse is due to persistence in getting unstuck and underway again.

Collected navigation image sets. We also periodically recorded image sequences at various sampling rates from the navigation cameras. These data sets are for future simulation testing of the navigator and for possible work in visual odometry. At this point we have tens of gigabytes of recorded images.

Tested fault recovery. We began a series of tests of the rover executive, the process that follow the mission plan and commands the rover and instruments. These tests involved intentionally crashing processes onboard, blinding sensors, slowing things down, and speeding other up, to induce the types of faults that occur over the duration of a long traverse. Zoë's rover executive is increasingly able to recover automatically from these types of faults.

Upcoming
• Public event with students visiting Zoe
• More autonomous traverse

Weather
Few clouds, windy 20-30kph with sustained gusts to 60kph, moderate 15-20C, 15% humidity

Quote of the Day
"Wagner Bog, Williams Drop, Tompkins Curve, Wettergreen Leap"