The great thing about artificial intelligence, and more specifically deep learning, is that it can work for just about anything if the model is trained properly. Deep learning models can run on relatively affordable hardware, which gives makers the ability to take advantage of their power for projects that can seem almost trivial. You could, for example, turn on a lamp with a simple light switch; or you could train a deep learning model to turn on the lights whenever you wave your arms around. James Bruton followed a similar approach when he built these Wolverine claws that automatically extend when an AI recognizes his intent to fight.
Obviously Bruton doesn’t plan on actually fighting, but this deep learning model is trained to recognize when he makes an angry face that signifies his desire to extend the claws. It would have been far easier and less expensive to extend the claws with a button press, but what is the fun in that? On a mechanical level, these claws work in a straightforward manner. The claws themselves, which were purchased online, are attached to a 3D-printed frame with a bungie cord wrapped around it. The bungie cord acts as a spring to store kinetic energy until the claws are extended. When triggered, a servo motor lifts up to release the kinetic energy and slide the claws out. There is no provision for retracting the claws, so they have to be “reloaded” manually.
The deep learning model for the artificial intelligence that controls the trigger is what makes this project interesting. That model is running on an NVIDIA Jetson Nano, which is a single-board computer that was created specifically for artificial intelligence projects like this. The Jetson Nano has a camera connected to it, which is mounted onto an arm that extends from Bruton’s chest. The camera is always pointed at his face in order to monitor his facial expression. The deep learning model, similar to the classic “Hot Dog or Not Hot Dog?” made famous by Silicon Valley, has been trained to differentiate between an angry grimace and every other possible facial expression. If it recognizes a grimace, the servo motor is triggered and Wolverine claws shoot out. This project may not have a practical purpose, but it does a great job of illustrating the versatility of deep learning models.