Figuring out what other people are thinking is tough, but figuring out
what a robot is thinking can be downright impossible. With no brains to
peer into, researchers have to work hard to dissect a bot's point of
view.
But inside a dark room at the Massachusetts Institute of Technology
(MIT), researchers are testing out their version of a system that lets
them see and analyze what autonomous robots, including flying drones, are "thinking." The scientists call the project the "measureable virtual reality" (MVR) system.
The virtual reality portion of the system is a simulated environment
that is projected onto the floor by a series of ceiling-mounted
projectors. The system is measurable because the robots moving around in
this virtual setting are equipped with motion capture sensors,
monitored by cameras, that let the researchers measure the movements of
the robots as they navigate their virtual environment.
The system is a "spin on conventional virtual reality that's designed
to visualize a robot's 'perceptions and understanding of the world,'"
Ali-akbar Agha-mohammadi, a post-doctoral associate at MIT's Aerospace
Controls Laboratory, said in a statement.
With the MVR system, the researchers can see the path a robot is going
to take to avoid an obstacle in its way, for example. In one experiment,
a person stood in the robot's path and the bot had to figure out the
best way to get around him.
A large pink dot appeared to follow the pacing man as he moved across
the room — a visual symbolization of the robot's perception of this
person in the environment, according to the researchers. As the robot
determined its next move, a series of lines, each representing a
possible route determined by the robot's algorithms, radiated across the
room in different patterns and colors, which shifted as the robot and
the man repositioned themselves. One, green line represented the optimal
route that the robot would eventually take.
"Normally, a robot may make some decision, but you can't quite tell
what's going on in its mind, why it's choosing a particular path,"
Agha-mohammadi said. "But if you can see the robot's plan projected on
the ground, you can connect what it perceives with what it does, to make
sense of its actions."
And understanding a robot's decision-making process is useful. For one
thing, it lets Agha-mohammadi and his colleagues improve the overall
function of autonomous robots, he said.
"As designers, when we can compare the robot’s perceptions
with how it acts, we can find bugs in our code much faster. For
example, if we fly a quadrotor [helicopter], and see something go wrong
in its mind, we can terminate the code before it hits the wall, or
breaks," Agha-mohammadi said.
This ability to improve an autonomous bot by taking cues from the
machine itself could have a big impact on the safety and efficiency of
new technologies like self-driving cars and package-delivery drones, the researchers said.
"There are a lot of problems that pop up because of uncertainty in the
real world, or hardware issues, and that's where our system can
significantly reduce the amount of effort spent by researchers to
pinpoint the causes," said Shayegan Omidshafiei, a graduate student at
MIT who helped develop the MVR system.
"Traditionally, physical and simulation systems were disjointed,"
Omidshafiei said. "You would have to go to the lowest level of your
code, break it down and try to figure out where the issues were coming
from. Now we have the capability to show low-level information in a
physical manner, so you don't have to go deep into your code, or
restructure your vision of how your algorithm works. You could see
applications where you might cut down a whole month of work into a few
days."
For now, the MVR system is only being used indoors, where it can test
autonomous robots in simulated rugged terrainbefore the machines
actually encounter the real world. The system could eventually let robot
designers test their bots in any environment they want during the
project's prototyping phase, Omidshafiei said.
"[The system] will enable faster prototyping and testing in
closer-to-reality environments," said Alberto Speranzon, a staff
research scientist at United Technologies Research Center, headquartered
in East Hartford, Connecticut, who was not involved in the research.
"It will also enable the testing of decision-making algorithms in
very harsh environments that are not readily available to scientists.
For example, with this technology, we could simulate clouds above an
environment monitored by a high-flying vehicle and have the video-
processing system dealing with semi-transparent obstructions."
0 comments:
Post a Comment