My father used to say that the most important thing in marriage was communication. He was Director of Operations for AT&T International, overseeing plants in Ireland, Saudi Arabia, the Netherlands, and several other countries. I think he learned about communication at work. The problem is, what are you supposed to communicate and to whom?

Today we learned a lot because eleven of us sat down for two hours and put together a plan for Boudreaux the robot. Being at MDRS provides an ideal opportunity for us to communicate—after all, the Mobile Agents engineers live and work in five different states on three coasts of the USA. But we also learned a bit about how we have failed to communicate, and how our designs could be improved.

As you see from the photos of the day, we sat around the table in the upper deck of the hab, facing a screen on which we projected a computer program called Compendium that we used for laying out a plan.

It took two hours to produce this plan for Boudreaux:

1. Move to waypoint #11 (a known location, which we recorded last year)
2. Deploy the wireless relay trailer
3. Pause (teleoperate to move away from the trailer)
4. Take a panorama
5. Move to waypoint X (to be determined by the geologists)
6. Pause (teleoperate to orient Boudreaux)
7. Perform a spiral scouting operation, with 3 meters between spirals and 3 meters between waypoints, for a maximum of 20 meters radius, recording communication waypoints every 2 meters and taking panoramas every 10 meters.
8. Pause (teleoperate optionally to view area beyond Pooh's Corner)
9. Take a panorama
10. Come home

To produce this plan, the various teams needed to learn some details about how the subsystems worked. For example, what happens if Boudreaux fails during the scouting operation (step 7)? The robot's control system sends a "failure" to the HabCom Agent, which presents this information to HabCom (the crew member) in MDRS. At this point, HabCom can tell Boudreaux to go to the next activity, to repeat the current activity, or to come home. Knowing this, we realized we should design the scouting operation to fail (setting a 20 meters radius that is sure to cause Boudreaux to encounter a hill or boulder obstacle). Then HabCom will tell Boudreaux (via the agent system) to go to the next activity. This maximizes the area that is scouted (recording waypoints automatically and taking pictures that we will use for planning the astronaut's EVA).

Today we came to understand subtleties for the first time in how the various levels of the Mobile Agents system (as it can be configured this year) interact. Together, we figured out how to exploit flexibilities in the existing Mobile Agents software and hardware to make a plan that is likely to succeed (hence, the multiple "pause and teleoperate" steps). The interacting levels are:

1. Science Work To Be Done
2. Agent System & Plan
3. Boudreaux, The Robot
4. Wireless Computer Network

I noticed a pattern: At each level, people would ask those in the level below, "How does this work?" And the levels below sometimes report back, "You can't do that." So we learned that the wireless network is ten times more powerful than what we've used before, and placing an antenna on top of the hab (instead of 2 meters off the ground) greatly extends our reach. But we don't know whether this base station will appear stronger than a relay that is inherently weaker, but closer to an astronaut. We learned how Boudreaux will search an area (in rows or a spiral) to map out the areas where a wireless signal exists. But we learned that Boudreaux will quit if an obstacle forces it to detour too far beyond its spiral path. We learned how to use Compendium to produce a plan (the 10 steps above). But HabCom can't tell the agent system to skip a step. Finally, we learned what areas were of interest to the geologists, and found that despite our original motivation in designing Mobile Agents, the geologists were forced to calculate GPS coordinates (to give us waypoint X, step 5).

I wondered if we could have communicated all of these methods, interactions, and limitations sooner. But as I discovered when we started this work three years ago, how do you know what to tell your teammates? How do they know what questions to ask? You really can't discover what needs to be communicated until you sit down and design something together. Indeed, we were eleven people from six different disciplines: geology, agent programming, speech processing, robotics, wireless networking, and model-based planning. Even if we had found the time to have a teleconference, would we have known enough about what we wanted to do to understand where our designs were too limited? Our best guide today was the aerial map of Pooh's Corner, by which we could all see how scouting would operate (the parameters mentioned in step 7), and those responsible for the scouting operation could realize what the wireless network below provided and what the agent system and scientists above really needed. That's why I have said what we do here is "empirical requirements analysis"—we figure out what to build by using prototype tools in a real work setting.

Yesterday I wrote about how I was interrupted less than every minute during a half-hour of frenzied work. Someone said to me today, "This shows how concentration and efficiency have to do with being able to 'turn off' your reactive behavior." But I don't think that accomplishing nine high-priority things in 30 minutes is inefficient! Just the opposite. By the same token, is a two hour meeting inefficient? Is doing this at MDRS and not months earlier inefficient? Well, if eleven people from six disciplines come to a common understanding about such a complex, layered system as Mobile Agents in just two hours, is that inefficient?

On the contrary, I think we showed today that working at MDRS is a great way to learn; the very setting and simulating planetary EVAs motivates communicating. Here we realize what questions to ask and what needs to be explained to others.

Sure, the change to Boudreaux's interface might have been communicated a day sooner. When I was told yesterday, "We need to integrate the ERA (Boudreaux) with Brahms (the agent system)" (which I recorded in yesterday's "Plans for Tomorrow" in my Check-In report), I should have asked, "What does that involve?" And I am sure that what Jeff would have said then would have caused Ron to raise his eyebrows.

So when NASA funds computer scientists and engineers like us to build a system like Mobile Agents, we're advancing knowledge about how to explore the moon or Mars. But more importantly, perhaps, we're learning how to work together across disciplines and different NASA centers. We're learning how to communicate.