The short answer is Yes! We don't use the term "AI" as a marketing gimmick. To our knowledge, we are first to build a pet laser product that actually uses Machine Learning (ML) techniques to automatically control the laser.
Felik contains a number of ML models that perform real-time inference locally on the device, without sending any of your data to the cloud. We use AI for "semantic segmentation" to separate tracked objects from the rest of the room, as well as for "classification" tasks - identifying specific objects for what they are (laser dot, pet, human). Additionally, AI controls the movements of the laser and decides where it should go based on predictions it makes about what your pet might do next, their history of movement, play area, restricted zones, and other parameters.
As with most ML models, the ones used in Felik aren't 100% accurate. For example, the classifier that detects the laser dot (so that it's not tracked as movement) is about 85% accurate. We had to strike a delicate balance between accuracy and making the classifiers small enough to work on an "edge" device - the quad-core 1.2GHz CPU with 512 MB of RAM. To do this we needed to carefully collect and label the training data, write a lot of custom feature extraction code, and optimize our neural networks to be as small and fast as possible while still providing accurate results.
Training Data Exampes For Object Classifier (Left: Positive, Right: Negative)