Machine learning works by finding patterns between inputs and outputs, causes and effects, or actions and outcomes. For example, imagine if thermostats didn’t exist and you had to manually turn on your air conditioner when you were hot, and turn it off when you were cold. A machine learning system could watch that pattern, and eventually come to the conclusion that you like the AC on when the temperature is above 72° F and off when it’s below 68° F. That simple process can be expanded to include far more complex tasks, which is how the oSense wearable can determine what you’re doing based on how you move your hand.
The oSense wearable device consists of six IMU (Inertial Measurement Unit) sensors that are monitored by an Arduino. Those IMUs are attached to the users fingers and the back of their hand. When they move their hand or fingers, that movement is picked up by the sensors, the data is collected by the Arduino, and then that data is analyzed by a machine learning algorithm running on a computer. The machine learning algorithm is then able to infer what the user is doing based on the sensor readings it has received.
Like any other machine learning system, oSense needed to be trained on those actions. That’s as simple as repeating the same tasks over and over again while telling the system what action is being performed. For example, drinking a glass of water. Eventually, the machine algorithm learns what sensor readings correspond to drinking a glass of water. What makes this interesting is how it can be used for evaluation and training. A professional calligrapher could write out a lesson while wearing the glove to train oSense, and then a student could repeat the lesson. oSense could then determine how closely the student followed the lesson, and which movements were different.