Today's Hardware Friday post is a different idea of how robot navigation can be made much simpler and easier using QR codes...
Getting a robot to know where it is and navigate to new locations is a tough problem that needs Bayesian particle filters, sophisticated machine vision, planning and ... or does it?
The localization problem is at the heart of robot navigation. If you want a robot to bring you a beer or a coffee then the robot needs to know where it is and where the coffee is. In short, the robot needs a map and it needs to know where it is on the map at any moment in time. A slightly more complicated problem is getting the robot to learn or construct the map as it works out where it is - the so called Simultaneous Localization and Mapping or SLAM problem.
Interesting though these problems are, there is another way of doing things. Most robots navigate around a "built" environment and there is no reason why we can't simply augment the environment so that the robot finds it easier to find its position. You can think up advanced technological ways of labeling the environment with RF beacons and so on, but a much simpler way is to use a real physical label.
QR codes, sometimes called robot vomit, might just be more useful than we thought. By sticking QR codes on walls, doors and other features, the environment can be signposted in a way that any robot equipped with a video camera can read. Of course, it isn't quite that simple in that the software has to detect the QR code sign, the robot can then move closer to read it and finally work out where it is.
An implementation of this idea by Aldebaran Robotics using the Nao proves that it could work with a full size robot.
... [Read the Rest]
Abstract—In this paper, we put forward a method for humanoid robot indoor localization and navigation, using 2D bar codes, running on embedded hardware. Our approach is based on camera image processing and the detection of 2D bar codes stuck to the walls. The particularities of our approach are: it works without odometry, fast localization (1s) that allows for kidnapping and falls, and it does not require the manual edition of a map (thus no extra computer is needed).
Results of an experimental study conducted with three different NAO robots in our office show that the proposed system is operational and usable in domestic environments. The NAO robots are able to navigate through a corridor of 5m in under 86s on average. Moreover using our system the robots manage to map and navigate a complex environment with multiple rooms that includes doors and furniture.
Taken together our results demonstrate that our navigation system could become a standard feature for our robots.
In this paper we propose to use bar code landmarks in order to provide an accurate indoor navigation capabilities to NAO humanoid robot. Experimental results obtained with three different robots show that the proposed system is fully operational and robust. Moreover the navigation speed is quite fast compared to previous work found in the literature and the proposed approach even works on robot with a huge drift. Future work will focus on the improvement of the landmark detection during movements and in different light conditions. Applications that could be based on our approach include: objects search in the user’s flat, going to the charging station, or game application like hide and seek.
This statement made me chuckle a little, "... During the different evaluation runs, humans users present at the office report that the head rotation of the robot head could be disturbing because it went over pi=4. A threshold on the head orientation should thus be envisioned...." So I guess Exorcist robo's are a no go...
Anyway, this sounds like an interesting robot navigation method. Of course we're not really ready to wallpaper our offices and homes in QR codes... yet anyway...