Machine learning technology has progressed dramatically in recent years, to the point where it is now practical for a wide range of everyday applications. There are a number of open source machine learning models available that anyone can use to add artificial intelligence to their projects, as well as hardware dedicated to running those models. Even so, machine learning is complex technology that can be really difficult to wrap your mind around and just setting up an environment for experimentation can be overwhelming. That’s why Google has a designed an affordable “teachable object” called Alto that makes machine learning approachable to the average maker.
Alto is a cute little robot-like device that was created specifically to introduce makers like you to machine learning. Just like Google Cardboard did for virtual reality, Alto makes machine learning approachable by using cheap materials to reduce cost and a streamlined toolchain to take the headache out of getting started. The Alto device is constructed from cardboard and cardstock, which can be cut from templates and formed by hand. The assembled device is a small box with a camera “eye,” two blocky arms, and a single button. That is obviously very limited, but the goal here is just to learn the basics of machine learning and not to build a robot that is going to do real work.
Inside of Alto’s cardboard enclosure is a Raspberry Pi Zero W single-board computer paired with a Coral USB Accelerator. The latter adds an Edge TPU that the Raspberry Pi can use to dramatically improve the performance and efficiency of machine learning models based on Google’s TensorFlow. Alto’s default machine learning model handles object recognition. It looks at objects through a Raspberry Pi camera and attempts to identify those using images it has already been trained on. If it does recognize an object, it will raise an arm to point at it. It can also learn to recognize new objects that you show to it.
That is neat and will certainly help you become acquainted with how machine learning models are trained, but Alto’s real potential comes from the fact that you can experiment with all of the other Coral examples that are available. Those range from monitoring a person’s movement in a video to voice recognition. With all of those examples to learn from, you can use Alto to gain an understanding of machine learning and eventually incorporate models into your own projects.