Since we started our Kickstarter campaign for MARK we have noticed a number of people wondering why do we have “bomb” class in common objects recognition pre-trained model. In this short blog post we’ll explain the reason for including this item and demonstrate a demo task that can be done with it – the patrol robot, searching the areas for dangerous objects.
The model’s dataset images for “bomb” class are downloaded from Google images and include pictures of cartoon-y timebomb with a clock in the front part, similar to what you see on the reference flashcard, included in MARK box.
For this demo we will also use MARK’s AprilTag detection function for implementing indoor navigation.
AprilTags are conceptually similar to QR Codes, in that they are a type of two-dimensional bar code. However, they are designed to encode far smaller data payloads (between 4 and 12 bits), allowing them to be detected more robustly and from longer ranges. Further, they are designed for high localization accuracy— you can compute the precise 3D position of the AprilTag with respect to the camera.The APRIL Robotics Laboratory at the University of Michigan
This function, coupled with neural network inference capabilities is what makes MARK stand out from a crowd of STEM education robots. When you’re designing a lesson for your students or a robotics competition task, with traditional robots you can only rely on line following for primitive navigation. With MARK you can use AprilTags, various items and flashcards to make tasks students can implement more interesting and somewhat closer to real-life applications.
In the sample demo we designer MARK starts at random point in room, where Area 1 Tag is visible. After reaching the vicinity of Area 1 Tag (vicinity is calculated by setting the AprilTag area threshold), MARK locates Area 2 Tag and proceeds in this direction. After reaching Area 2, the robot starts sweep scan of the area. If a “bomb” object is recognized on the ground, it then sounds the alarm and blinks headlights red. For “bomb” object both flashcard and a mock “bomb” can be used. We used adversarial attack approach to design an object that is recognized as “bomb” by the model.
In the last demo, just for a little bit of fun, we changed the code for MARK to ram the “bomb” when it is recognized. Rest assured, no robots were harmed during making this video.
Stay tuned for more articles from us and updates on MARK Kickstarter campaign.
BLOG UPDATE JULY 2020: You can now find MARK on our online shop
For more information on Grove Zero series, Codecraft and other hardware for makers and STEM educators, visit our website, https://tinkergen.com/.