New Navigation Method Helps Robots Locate the Front Door Without Advanced Mapping
Sat, April 10, 2021

New Navigation Method Helps Robots Locate the Front Door Without Advanced Mapping

The machine that will deliver a package is trained to recognize the lead from the sidewalk to help the robot recognize the front door. / Photo by: Andriy Popov via 123rf

 

When it comes to robotic navigation, the standard approaches involve advanced mapping of the area to avoid unsafe conditions and collisions. This requires the roboticists to use algorithms that can guide the machine in a GPS-coordinated environment and achieve a certain goal. This approach makes sense if the robot will be used in exploring certain environments, like in an obstacle course or a building, but not in terms of last-mile delivery.

Navigation Method for Last-Mile Delivery

This inspired Massachusetts Institute of Technology researchers to develop a new method that could help robots find the front door without the standard approach to robot navigation. Instead, the robot will use clues found in its environment so that it can plan the route towards its destination. Such a destination can be described to the robot in semantic terms like “garage” or “front door” so that the robot will not rely on coordinates.

The MIT team realized that mapping every single neighborhood for the robot delivery zone would be too challenging to scale, especially if it involves the whole city or if some houses change their exteriors with every season. Such an approach to navigation may also bring issues of privacy and security on the part of the homeowners. So, the MIT engineers preferred a navigation system that relies on clues. 

What Are These Clues?

For instance, if the machine is instructed to deliver a package into a certain home, it will begin to see the road and then the driveway. The MIT team shared on their website that they have already trained the technology to recognize the lead from the sidewalk to help the robot recognize the front door.

The MIT wrote that their new navigation technique will help reduce the time that the machine uses to explore a house before it can finally determine its target. MIT Department of Mechanical Engineering’s graduate student Michael Everett, who was one of the researchers behind the navigation system, said that they no longer need to gather the map of each building or property in the city that the robot will be visiting or need to visit. Rather, the robot can just be dropped off at the driveway and, on its own, will locate the front door and deliver the package.

The results of their research will be presented by Everett at the International Conference on Intelligent Robots and Systems. The co-authors of the study were Ford Motor Company’s Justin Miller and MIT”s professor of astronautics and aeronautics Jonathan How. Their work has been considered as a finalist for the Best Paper for Cognitive Robots.

Using Semantic Technique and Algorithm

Everett says that robots now have the ability to have “a sense of what things are” and it can be done in real-time. Prior to their work, other researchers had already introduced semantic and natural language to robot systems by training them to recognize semantic labels and objects. Then, it would be visually processed by the robot. For example, robots would visually process the door as a door and not just a rectangular or solid obstacle on its way.

Everett, Miller, and How used the same semantic technique for their robotic navigation approach. They believe that it leveraged the pre-existing algorithms that take out features from the visual data and then create a new map from the same scene but represented to the robot as context and semantic clues.

They also used an algorithm called Simultaneous Localization and Mapping or semantic SLAM. In the past, semantic algorithms would allow robots to determine objects and map them in their environment but could not help them in making decisions in real-time while also navigating. With their new navigation system, the Simultaneous Localization and Mapping algorithm enables the robot to make real-time decisions while it is navigating even in a new environment. It can just identify the most efficient route so that it can reach its semantic destination, like the front door.

 

Other researchers had already introduced semantic and natural language to robot systems by training them to recognize semantic labels and objects. / Photo by: arrow via 123rf

 

Cost-to-Go Estimator Algorithm

The researchers likewise developed an algorithm called the cost-to-go estimator. It helps convert the semantic map generated by the SLAM algorithm into another map (second) that represents the likelihood of a given location.  Everett says their algorithm was inspired by a class of vision and graphics called image to image translation.

Delivery Robots: Market Size

Database company Statista shared the market size of delivery robots worldwide in 2018 and its prediction for 2024, expressed in millions of US dollars. Globally, the market size of delivery robots in 2018 was valued at 11.9 million and is expected to reach 84 million in 2024. 

In a 2019 report by the market research company Markets and Markets, it was revealed that the delivery robot market will grow at a CAGR of 19.15 percent by 2024. Currently, the cost of every last-mile delivery is $1.60 via human drivers. By utilizing autonomous delivery machines, the cost can be scaled down to $0.06 for every last-mile delivery.