Inspiration: Recognizing the demand in our market for automation of human tasks in indoor environments and our group having the experience with building and training neural networks and working with hardware, we decided to functionalize a fundamental feature that mobile robots operating in indoor environments would need - door opening recognition.
What it does: After collecting our own data using an ultrasonic sensor we built and trained a neural network model that can be uploaded to a microprocessor to allow a robot with an ultrasonic sensor to identify whether a door in front of it is open or closed after scanning it. This has been tested by uploading data from a scanner and resulted in 92% accuracy of predictions. Challenges: Working in different time zones, formatting collected data, visualizing collected data, converting the model into a format that can be used by the robot.
What I am proud of: We are proud that we got to accomplish so much in such a short timeframe with so many obstacles before us!
What I learned: -Workflow on github as a member of a team. -Neural Networks and how they work -Tensorflow
What's Next: We will keep working on loading the model onto the esp32 microprocessor and start working on real-world applications of our model as it is uploaded to the robot.
Log in or sign up for Devpost to join the conversation.