...in the international history of the space age, only five of 13 attempts to land on Mars have succeeded.
Check out this great picture of the part of the landing sequence where Phoenix descends into the Martian atmosphere with its parachute.
A crucial mission sequence is called a critical sequence. In a critical sequence you get one and only one chance to to execute the sequence right, with a failed sequence implying a failed mission. Critical sequences include landings (like the one that Phoenix executed flawlessly), orbit insertions (such as Cassini inserting itself into Saturn's orbit), and some flybys (such as Stardust's flyby of Comet Wild 2 to collect and return comet dust).
Talk of critical sequences reminds me of our work on the Remote Agent. I was at NASA through much of the 1990s working as a research scientist in an Artificial Intelligence (AI) research lab. Some time in 1995 the NASA Administrator, Dan Goldin, gave an inspiring speech. In it he outlined a vision that would allow NASA to significantly step up space exploration in spite of shrinking budgets. The key, he said, was to develop new technologies in such diverse areas as propulsion, MEMS, and autonomy. And he established the New Millennium Program to carry out his vision.
We (i.e., a group of AI researchers) were energized by this speech and decided to do something about it. Specifically, we thought that we could build a Remote Agent---an on-board autonomous agent to provide high-level control to achieve mission objectives. We pitched our idea to spacecraft engineers at JPL. Naturally, the spacecraft engineers were skeptical---after all, what did a bunch of AI researchers know about spacecraft autonomy? But they were open minded---they said they'd believe us if we could show them how the Remote Agent would handle Cassini's critical sequence (Saturn orbit insertion) in a simulated environment. And to make sure that the simulated environment didn't abstract away the hard parts of the critical sequence, they assigned an accomplished spacecraft engineer to lead the project.
We worked hard for the next 5 months or so, leading up to a demonstration of the system to various senior engineers at JPL. Everything was going smoothly, when suddenly an unexpected bug hit us (unexpected in the sense that it was not part of the demonstration script, though we were aware of this bug in our earlier testing but hadn't had the chance to track it down and fix it). The bug was in the communication between the Remote Agent and the underlying spacecraft simulation, and it manifested itself as a failure of the main rocket engines just as they were being turned on to slow down the spacecraft in preparation for orbit insertion!
The Remote Agent responded beautifully to this unexpected turn of events! It noticed that the main rocket engine had failed, switched to the back up engine, replanned the insertion sequence, and successfully inserted the spacecraft into orbit around Saturn (all in simulation of course---the real Cassini mission didn't use the Remote Agent!).
The success of this demonstration led directly to doing this for real on Deep Space 1, the first of the New Millennium missions. On May 17, 1999, the Remote Agent took control of Deep Space 1 to start the first of two on-board autonomy experiments. Being in mission control during these experiments was one of the most thrilling moments of my professional life!
You can read all about the Remote Agent here and about the actual experiments here. No discussion of a space mission is complete without a photograph of the spacecraft. I wanted to include a photograph of me with Deep Space 1 (in the testing bay at JPL). Unfortunately, I couldn't locate that photograph :-( So instead, you'll have to make do with the following poster. I have a big version of this poster at home---a parting gift from my NASA colleagues when I left NASA.
No comments:
Post a Comment