NASA's roving rover uses artificial intelligence to steer round rocksSteve BushNASA is looking at neural networks and fuzzy logic to control its next generation ofplanet rovers.
"We want to tell the robot to think about any obstacle it encounters just as anastronaut in the same situation would do," said Dr Ayanna Howard, artificialintelligence scientist at the Jet Propulsion Laboratory in California. "Our job is tohelp the robot think in more logical terms about turning left or right, not just by howmany degrees."
All current NASA space craft and rovers (except Deep Space 1 which is used forartificial intelligence research) are controlled by a combination of pre-programmed andradio commands.
The ultimate goal for Howard and colleagues is "putting a robot on Mars andwalking away, leaving it to work without direct human interaction," she said.
NASA's prototype system is tailored to operate in sand-and-rocks environments,choosing its own route to operator-selected locations. small rock large rock large rock roughness FEW FEW FEW MANY FEW ROUGH MANY MANY ROUGH MANY CLOSE ROCKY
En-route it takes into account of: terrain roughness, terrain slope, surfacediscontinuities (ditches) and local objects.
Behavioural models digest processed image data and vie to steer the rover through a final analysis block.
This involves initially identifying linear features to locate edges in the landscape.These edges allow the local terrain to be separated from distant objects, as well asspotting potential ditches and cliff edges.
Then the local terrain is analysed for rocks, the size (‘small' or‘large') of the rocks and their location relative to each other.
Analysed terrain data feeds into a layer of behavioural models, one for the main aim -to get to point X, one for terrain based navigation - finding likely routes, and one forlocal collision avoidance.
In particular, the route-finder rules-out areas likely to be too rocky and looks forsmooth paths among the rocks in remaining parts of the local terrain.
The behavioural models then vie to persuade a final analysis block where to drive therover.
At the moment, the whole thing runs on a 333MHz Pentium II Linux-based computer in aCompact PCI chassis. Data comes from six high-mounted CMOS NTSC video cameras through two4-input framegrabber boards. The whole lot sits on a prototype rover with a trailer.