Localization, navigation and safety

Localization and navigation are critical requirements for a high-speed device such as the BallBOPPer™. The BallBOPPer™ currently handles localization and navigation in a couple of different ways.

From the beginning, the BallBOPPer has used Vision AI that is trained for landmark recognition for navigation. The court landmarks we use include the net, the net posts, the center strap on the net, the various intersections of the court lines, the surrounding fences and fence posts.

The main problem we have with this approach is when the court landmarks are less obvious. One common one is with side-by-side courts that lack fencing or any other marker between the courts. The potential for collisions with players on the other courts is real.

In addition, a Vision AI system has to deal with widely different court configurations, including different surfaces inside the fences, as well as sun and weather variations.

During training, and as a safety measure during testing, we use an Ultra Wide Band (UWB) system which, because it is doing time-of-flight calculations, is accurate to about 6 inches and is very fast.

This system works well and has proven reliable, but it is high maintenance, the beacons have to be precisely positioned, and each has its own set of batteries. But the main issue is that with the units we currently use, the Decawave DW-1000 units, the range is limited to about 40 feet. This requires setting-up 8 UWB beacons around the BallBOPPer end of the court. It requires a fair amount of effort, and we therefore use UWB for safety when testing, but we don't want users to have to use it when training with the BallBOPPer.

Besides its accuracy, one of the big advantages of the UWB system is that you can clearly demarcate the sides of the court when there aren't any side partitions by placing the side beacons along the line where you want the partition to be. The BallBOPPer™ is then confined to traveling inside the beacons.

There are a number of other technologies we have experimented with. Real Time Kinematics (RTK), which is used by a number of robo lawn mowers, provides the same kind of accuracy as UWB, but it requires a clear view of a substantial number of GPS satellites, and it does not work indoors.

Lidar, radar and stereo depth cameras all provide useful information, but add expense and complexity, and still require other technologies to fill in their gaps.

The one system that has the potential of covering all the bases is still the Vision AI system we have been working with since the beginning.

We are in the process of running experiments involving retraining the BallBOPPer™ Vision AI system to use something like AprilTags in combination with the court landmarks.

AprilTags are high‑contrast, square fiducial markers used in robotics competitions for precise, camera‑based localization.

They are like QR codes engineered specifically for robots: easy to detect, easy to decode, and mathematically optimized to provide 6‑DoF pose relative to the tag.

By putting tags on small stands that can be placed around the perimeter of the court, particularly along sidelines without partitions, we hope to achieve the kind of flexibility we have with the UWB beacons, but without the batteries, and without the need to place them in precise positions.