Autonomous robots already roam sidewalks across the US, delivering packages and food to customers’ doorsteps—assuming they can find the front door.
Traditionally, robotic navigation requires mapping an area in advance, then using algorithms to guide the device toward specific GPS coordinates.
But MIT has a better idea: Engineers developed a new approach that enables a robot to use context clues to plan the route to its destination—described in general terms like “front door” or “garage.”
This technique, according to the institute, greatly reduces the amount of time the bot needs to explore a property and identify its target. Plus, it doesn’t rely on maps of specific residences.
“We wouldn’t want to have to make a map of every building that we’d need to visit,” Michael Everett, a graduate student in MIT’s Department of Mechanical Engineering, said in a statement. “With this technique, we hope to drop a robot at the end of any driveway and have it find a door.”
The recent introduction of natural language to machine-learning systems lets robots recognize objects by their semantic label: A door is a door, not just a solid, rectangular obstacle.
“Now we have an ability to give robots a sense of what things are, in real time,” Everett explained.
In collaboration with MIT professor Jonathan How and Justin Miller of the Ford Motor Company, Everett used the semantic SLAM (Simultaneous Localization and Mapping) algorithm to build a map of the robot’s environment as it moved, using semantic labels and a depth image.
To speed up the robot’s path-planning, the team developed a “cost-to-go” algorithm, which converts a semantic map into something representing locations far from and close to any given goal.
For instance, the sidewalk—coded in yellow in a semantic map—might be translated by the cost-to-go algorithm as a darker region in the new map. While a driveway, which gets progressively lighter as it approaches the front door, may be the lightest region in the new map.
Researchers trained their new algorithm on satellite images from Bing Maps, containing 77 houses from one urban and three suburban neighborhoods.
They also applied masks to each image to mimic the partial view a robot’s camera would likely have as it traverses a yard.
“Part of the trick to our approach was [giving the system] lots of partial images,” according to How. “So it really had to figure out how all this stuff was interrelated. That’s part of what makes this work robustly.”
In a simulation, the group’s new cost-to-go technique found the front door 189 percent faster than classical navigation algorithms.
More on Geek.com:
- Amazon Launches Scout Delivery Robots in Washington
- MIT Engineers Develop ‘Blackest Black’ Material
- MIT’s Color-Changing Ink Lets You Customize Shoes, Phone Cases
from Geek.com https://ift.tt/2JSw2f1
via IFTTT
0 comments:
Post a Comment