MIT Engineers Teach Delivery Robots To Self-Navigate To The Door

A prospective of delivery robots dropping a meal-kit, a package or other takeout doesn’t seem too far away from your expectation. However, if you want your stuff to be shipped at the door, the robots will need to find the door in advance. Now, engineers from MIT have teamed up with partners from Ford Motor Company to ease the process, counting on a technique that doesn’t require robots to navigate via maps but clues.

In most cases, a location that has been detailed on the map is a vital requirement for robots to navigate themselves to that place. However, it’ll become much more challenging if we do mapping on a large scale, not mention a lot of concerns over security and privacy issues. The method of using clues now allows robots to level up their abilities when there’s no specific data of the area at all. In general, some clues can be seen as “front door” or “garage.”

According to MIT engineers, they’ve likely trained the robot to learn that a front door can be found on a sidewalk, and a sidewalk often comes from a driveway. Although every corner in the world doesn’t appear to be a copy or look-alike version of another, there are still some common stereotypes.

Along with using pre-existing algorithms that connect visual data and scene identifications with semantic clues, engineers will not only build a map at the area the robot moved around but also allow it to choose the most efficient way to arrive at the destination. As scheduled, these new findings will be on the agenda at an international conference this week.