PITTSBURGH — Autonomous navigation software developed by researchers at Carnegie Mellon University and NASA’s Jet Propulsion Laboratory (JPL) drove the NASA Mars Rover “Opportunity” last week in the software’s first live test on the Red Planet.
Preliminary data from the test run, which was initiated Feb. 7, indicates it was successful. Opportunity, operating near the rim of Victoria Crater, was traversing an area that mission managers had made certain was without obstacles for this initial test. But they programmed a “virtual keepout zone” for Opportunity to maneuver around, and initial findings suggest the rover veered around this simulated obstacle as anticipated.
“The early indications suggest that the software is operating as designed,” said Tony Stentz, research professor at Carnegie Mellon’s Robotics Institute and associate director of the National Robotics Engineering Center. “Much more work and testing remains to be done, but we are thrilled to see our software operating on Mars and we believe it will ultimately expand the capabilities of this and future planetary rovers.”
The software is designed to not only help Opportunity avoid obstacles but also to maneuver its way out of the dead-end cul-de-sacs it sometimes encounters in complex terrain.
In previous tests, which began in November, the navigation software acted as a “backseat driver,” producing planned paths for the Rover rover but not actually controlling it. In last week’s test, the software moved from the backseat to the driver’s seat for the first time. Additional tests are anticipated and eventually the new software could become a control mode that is routinely used for the rovers, Stentz said.
Opportunity and its twin, “Spirit,” were designed to operate for only a few months but have proven surprisingly resilient. Last month, they began their fourth year of operation on Mars. Prior to the software test last week, JPL officials announced that Opportunity had extended its Martian driving record to 10 kilometers, or 6.2 miles — 17 times the distance it was to cover in its mission.
The longevity of the rovers has enabled not only continued exploration of Mars, but also the testing of new capabilities. Revised flight software for the enhanced autonomous navigation, as well as several other new skills, was uploaded to the rovers last year.
Both rovers can already be operated in an autonomous mode, allowing them to turn to avoid an obstacle or hazard. But that software allows the rovers to think only one step ahead: if the rover encounters another obstacle after turning, it turns again. This behavior doesn’t work well if the rover has driven into a cul-de-sac, Stentz said. Unable to escape, the rover can end up looping back on itself until human operators can manually drive the rover out of the dead end.
By contrast, the new software builds a large-scale map of the terrain it encounters, giving the robot a memory of how it got where it is and the ability to retrace its steps or plot a new way out if necessary.
The software is based on a program called Field D* created by Stentz and his student, Dave Ferguson, under sponsorship of the Army Research Laboratory. It has been used to control a number of mobile robots at Carnegie Mellon, including “Crusher,” the experimental unmanned ground combat vehicle being developed for the Defense Advanced Research Projects Agency (DARPA). It is also used in General Dynamics’ experimental unmanned vehicle (XUV). An earlier version of the software, called D*, dates to the early 1990s and the DARPA-sponsored NavLab 2, a Humvee with off-road autonomous navigation capability.
One strength of Field D* is that it is computationally efficient, Stentz said. Crusher, for instance, can produce new path plans several times a second. But the computers onboard Spirit and Opportunity are only about one-hundredth as fast as a typical desktop computer, far less powerful than the processors onboard Crusher. So in spring 2005, Stentz and Ferguson, now a research scientist at Intel Research Pittsburgh, began adapting Field D* to run on a weaker processor. They delivered it to JPL, where Carnegie Mellon alums Mark Maimone and Joseph Carsten, along with technical staff member Arturo Rankin, integrated it into the Mars Exploration Rover software.
After the software was demonstrated on a robot in JPL’s “Mars sandbox” simulation site, the software was uploaded to the Mars rovers last summer. JPL, a division of the California Institute of Technology, Pasadena, manages the Mars Exploration Rover Mission for the NASA Science Mission Directorate in Washington, D.C. For more information, visit the mission home page at http://marsrovers.jpl.nasa.gov/home/index.html