Georgia Tech tests unmanned systems
The Georgia Tech Research Institute (GTRI) will spend the next year developing a road map designed to improve the testing and evaluation of unmanned and autonomous systems for the Office of the Secretary of Defense.
This project brings together concept development and system evaluation, focusing on new ways of testing unmanned aerial vehicles and unmanned systems that operate on land, on the sea and underwater. The focus is chiefly on figuring out how to test the next generation of unmanned systems that will be able to operate much more autonomously.
“The next generation of these vehicles is expected to be much more autonomous and intelligent, with the ability to make their own decisions,” said Lora Weiss, a GTRI principal research engineer.
Known as the Roadmap Development and Technology Insertion Plan, the one-year, $430,000 contract announced in July is funded through the Army at White Sands Missile Range. So far no unmanned vehicles embody the kind of on board intelligence Weiss and her collaborators are concerned with. Instead, their computer power is mostly used to simplify work for the human operator. But someone still must determine where the vehicle goes and what it does, either with a joystick control or by directing it to follow a flight path — or the equivalent for land and sea vehicles — as a connect-the-dots series of predefined waypoints, Weiss said.
Prototypes of more autonomous systems exist as research projects and have been put to the test in settings such as the Defense Advanced Research Projects Agency Grand Challenge robotic road races in deserts and the 2007 DARPA Urban Challenge race through a simulated city setting. To finish a task without the guidance of a human operator, a robotic vehicle needs to be able to work through nondeterministic scenarios — another way of saying it needs to be able to expect the unexpected.
Deploying systems that are smart enough to accept general directions will require figuring out ways to test whether those systems are capable of making the right decisions without explicit guidance, Weiss said.
First, researchers will need to figure out how to devise a wide enough variety of unexpected scenarios that simulate the real world. Even if a vehicle makes the correct decision in a specific scenario, researchers don’t necessarily know whether that’s because of correct logic in the operating software or whether the device stumbled randomly onto the right choice and might get it wrong next time. As a practical matter, particularly with some of the small unmanned vehicles, there will also be a limit to how much instrumentation gear can be loaded on board during testing and how much decision-making data can be logged during operation.
Giving unmanned vehicles more autonomy should make them more productive, allowing one warfighter to supervise dozens of vehicles, Weiss said.
Another challenge will be testing scenarios in which several unmanned vehicles collaborate to complete a given task, said Rusty Roberts, a principal research engineer who oversees all of GTRI’s test and evaluation programs.
At first, a team of unmanned vehicles would likely summon a human observer to review its observations rather than taking lethal action autonomously. The interactions among unmanned systems will be another element researchers need to test, Roberts said.
In this project, the GTRI researchers will produce a road map of ideas for autonomous unmanned system testing, Weiss said. “Then we expect to start building some prototypes,” she said.