Aerospace Technology Development

Smart Software "Learns" Precise Docking

DOCKING ONE SPACECRAFT WITH ANOTHER will be much easier, thanks to smart computer software being developed at NASA's Ames Research Center. The neural net software will "learn" the motion behavior of a spacecraft as it flies, so it will not undershoot or overshoot docking targets.

The Ames "neurocontroller" will be able to automatically dock a spacecraft of unknown mass with another by flying the spacecraft short distances in different directions and learning the handling characteristics of the craft, according to Dr. Robert W. Mah, an Ames research scientist. "The neurocontroller automatically enables precise, safe docking. This software learns similar to the way human beings learn—by experience, handling the spacecraft," Mah said. The neural net software used in the controller is similar to that used to automatically focus home camcorders.

"Current methods used to dock spacecraft will be improved to make docking easier for astronauts," Mah said. "Docking a spacecraft by manual joystick control depends on the skill of the operator."

Manual docking can be slower than desired, and in some cases precious fuel can be wasted. Conventional software has been used to automatically dock spacecraft, when the spacecraft mass properties are known, according to Mah. "But conventional automated docking doesn't work well when the exact mass is unknown or changing robot arm positions alter spacecraft flight characteristics," Mah said.

A future "worker bee" spacecraft, which would be used during construction in space, might grab a construction part and rotate, Mah said. "The arm and the part held by the spacecraft have a tendency to keep rotating," he added. In contrast, the same spacecraft equipped with the neurocontroller would immediately learn the new "feel" of the way the spacecraft rotates in space while firing its impulse jets. The smart software would then precisely slow the spacecraft's rotation by "burning" the jets for just the right amount of time.

The Ames neurocontroller has not yet been tested in space. "We hope to do space tests eventually, but first we need to fine-tune the software in the lab, making more realistic simulations with test vehicles floating on a cushion of air over a special granite table," Mah concluded.

The neural net software is a form of artificial intelligence. It can be used in the medical field for neurosurgery and for breast and prostate cancer surgery to identify tissues and tumors with less damage. The software has the ability to "learn" the characteristics of a system or different kinds of tissues, such as brain or tumor tissue. In the aircraft industry, neural net software could be used to safely land damaged planes and could be an effective tool for the next generation of flight controllers.

For more information, contact Dr. Robert Mah at Ames Research Center.
Call: 650/604-6044, Fax: 650/604-3594, E-mail: mah@mail.arc.nasa.gov
Please mention you read about it in Innovation.

Neural net software programs that are smart enough to "learn" the way humans learn may automatically dock spacecraft more precisely and safely.