Mobile delivery robots are being tested by a variety of companies as an alternative to standard package delivery or even drone delivery, Now, researchers at MIT have developed a navigation method that does not require mapping an area in advance.
The MIT method enables a robot to use clues in its environment to plan out a route to its destination, which could be anything from a front door or a garage, rather than just coordinates on a map. If a robot needs to deliver a package to a front door, for example, it might start on the road and then find a sidewalk that will lead it to the front door.
MIT said the technique reduces the time a robot has to spend exploring a property before identifying its target and does not need to 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,” said Michael Everett, a graduate student in MIT’s Department of Mechanical Engineering. “With this technique, we hope to drop a robot at the end of any driveway and have it find a door.”
The goal is to give robots a sense of what things are in real time using semantic techniques as a springboard for navigation approaches leveraging algorithms that extract features from visual data to generate new maps of the same scene. These semantic labels can be processed by robots as a door and not just simply as a rectangular obstacle, MIT said.
An algorithm was used to build up a map of the environment as the robot moved around using semantic labels of each object and depth image. These semantic algorithms allowed a robot in testing to make decisions in the moment while navigating a new environment — taking the most efficient path to the front door.
“Before, exploring was just, plop a robot down and say ‘go,’ and it will move around and eventually get there, but it will be slow,” said Jonathan How, professor of aeronautics and astronautics at MIT.
In training the robot, the system converted a semantic map into the most efficient path, following lighter regions on the map, to the end goal. In each satellite image, the team assigned semantic labels and colors to context features in a typical home such as grey for a front door, blue for a driveway and green for a hedge. This was then applied to the robot’s camera so it could search for a path in real time to a house it had never been to before.