Researchers in Artificial Intelligence have always dreamed of building the artificial person. Like Commander Data on Star Trek or the Bishop robot in the movie Aliens. But up until 1986, the best they could do were trashcan-like mobile platforms that carefully looked at the world (with cameras and sonars), analyzed what was going on, figured out what to do next, and then did it. It made for robots that did their jobs at a snail's pace, sometimes standing motionless in place for many minutes.

Then in 1986, a new idea emerged. Motivated by the way animals behave and the way humans carry out the routine activities of ordinary life, a new band of roboteers designed robots that could move about the world very quickly without running into things or falling down stairs. These robots were able to sense the world and act (or react) to what they saw many times a second. Now we had robots that could travel down hallways without hitting people, track moving objects, and even open doors, all at the same speeds as humans.

The problem was, because they only did one or two things really well, these robots would never be butlers or chauffeurs or could never search for drowned swimmers or repair satellites. So a small set of researchers began devising ways to layer intelligence over these agile, fast moving agents. TRACLabs researchers were leaders in this effort. We layered intelligence first by teaching the robots about routines, or recipes for action. For example, to fetch a cup for Linda from a table in the conference room, a simple recipe might be to first navigate to the conference room, then locate the cup, then move to the cup, then pick it up, then navigate to where Linda is. Since the post-1986 robots could navigate through office buildings and perform pick-and-place tasks with ease, these robots could execute these new action recipes with the same speed and agility.

But while our now-smarter robots could fetch a can of soup from a supermarket shelf, they couldn't go grocery shopping. So we layered another level of intelligence on our robots -- planning and scheduling. AI planning techniques were used on those pre-1986 robots to try to imbue them with a human-level of intelligence. But the planning routines of those early robots had to plan every wheel turn and every joint motion. Now, with our layered intelligence robots, the planning routines only have to plan down to the action recipes in the robot's first layer of intelligence. So our robots can now theoretically plan their whole day.

Over the past 15 years, TRACLAbs has used its 3T robot intelligence software to program robots to search for, find and recognize people, to recognize a user's gestures to carry out tasks, to perform inspection tasks for the space station, to hunt for mines underwater, and to form teams with humans to carry out repair and replacement tasks on earth or in space.

But there's more. It turns out that we can layer intelligence on any computer controlled machines, even ones that don't move or have arms, like the machines used to run a nuclear power plant or process waste water or recycle breathing air. While the reaction times are a lot slower for these "immobots", the layered intelligence principle applies. So for the past ten years, TRACLabs has also been busy developing intelligent control systems for all of JSC's advanced life support developments, like biological water processors, water distillation systems, oxygen generation systems and CO2 recovery systems. The results of several of these efforts were used in human rated tests, including one with four humans living and working in a NASA biosphere for three months.

We have also extended our approach to intelligent robotics to include graphical and natural language interfaces. So users of TRACLabs robots can motion-to, speak-to or just point-and-click the robots into action.

Robot All Stars

Nam enim diam, scelerisque eu
mattis a, volutpat vitae, lectus. Aenean faucibus magna et leo. Proin neque orci, condimentum sit amet, luctus id, dictum a, nibh. Nam vestibulum.