In the future, autonomous cars will not only take people to and from destinations, but they will also deliver goods without the need for drivers to operate the vehicles. The catch is that no driver means there will be no one to take the packages to the door. Many see robots as the answer to that dilemma.
The trick is helping the robots to find the door. A standard approach to this issue requires mapping an area ahead of time. After mapping the area, the robots can use a GPS coordinates on the map to find where they are going. The challenge for using this technique for last-mile delivery is that it would require a map to be made for every neighborhood within the delivery zone for the bot.
MIT has a new method of navigation that doesn’t require mapping the area in advance. The approach allows the robot to use clues in the environment to plan out the route to its destination. When a robot is trying to deliver packages to the front door, it would start on the road and look for the driveway, where it would recognize that likely leads to a sidewalk, which in turn leads to a front door.
The technique would reduce the time the bot spends exploring a property before finding its target. The team used an algorithm to build up a map of the environment as the robot moves. The method uses an algorithm called semantic SLAM (Simultaneous Localization and Mapping).
The team sped up the route-planning using a semantic, context-colored world. They also developed a “cost-to-go” estimator that converts the semantic maps created by the preexisting SLAM algorithms into a second map indicating the likelihood of a given location being close to the goal.