An introduction to Machine Learning using LEGO
summary
A LEGO Mindstorms car with a phone-mounted camera, a custom app for data collection and training, and steering angle as the label is trained to follow a black tape line. Training uses gradient descent on a linear model (then adding squared pixel features), 500 iterations and varied learning rates (0.001, 1, 2.5) to find convergence. Data came from ~20 minutes of driving using a single pixel row as features; the model mostly works but needs more/better data and faster hardware to handle sharp turns.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:52Project challenges
- 6:12Machine learning basics
- 12:19Training algorithms
- 18:30Data collection setup
- 30:48Model training and tuning
- 37:01Initial testing
- 44:18Refining the model
- 51:56Final conclusions
related talks
58mAn introduction to Machine Learning using LEGO
43mLessons Learned Using Machine Learning in Java
19mDeep Learning is Just LEGO: & Other Hands on Machine Learning Activities
48mLessons learned using Machine Learning in Java
51mMachine Learning Exposed Introduction to Machine Learning
58mThe Modern Trolley Problem - Responsible AI Principles
01h 25mHow to Build Your Own Self Driving Toy Car
49mMachine Learning Exposed: Introduction to Machine Learning