How to Build Your Own Self Driving Toy Car
summary
Six months of work produce a drivable 12" yellow Ferrari toy car wired with a Raspberry Pi camera and controlled by models built in OpenCV and TensorFlow. Training uses about 10 GB of footage of 320×240 color frames (~200k pixels each); a simple single hidden layer (32 neurons) network surprisingly outperforms deeper CNNs for this setup. Haar‑cascade stop‑sign detection runs fast enough on the Pi, but models run on a laptop or AWS GPU (≈10× speedup) because the Pi’s inference latency is too high for reliable real‑time control.
AI-written summary · about
chapters (33)
- 0:00How to build your own Self-Driving Car
- 0:51About Me
- 3:16Why I did this project
- 6:24Materials
- 10:26Internals
- 13:49Training Track
- 19:26What I’ll demo today
- 20:19Car camera demo
- 21:46Tech Stack
- 22:44Raspberry Pi
- 24:09Raspberry Pi vs Arduino
- 28:58Hello World for the Pi: LED light
- 29:41Gradually added more complexity
- 30:16Pro Tip: Don’t inhale lead
- 31:16OpenCV
- 32:32Haar Cascades
- 37:50Stop sign detector
- 42:35TensorFlow
- 44:43Machine Learning 101
- 46:49How a Neural Network “Sees”
- 50:39Convoluted Neural Networks
- 52:06Deep Learning
- 52:53Standard Neural Network
- 56:20Tips for tuning Neural Networks
- 59:11Tensor Board
- 1:00:44GPU Machine Learning
- 1:02:50GPUs now commonplace
- 1:04:28NVIDIA stock price
- 1:05:23Machine Learning takeaways
- 1:09:01Car and hardware tips
- 1:10:58To Do
- 1:12:18That’s it!
- 1:13:00Question & Answer
related talks
35mDingoCar: Perception module Small scale, self-driving car
33mSuper Donkey Cars
47mUnlocking Data-Driven Apps with TensorFlow
42mAI and Self-Driving Cars
58mAn introduction to Machine Learning using LEGO
40mTeaching a Car to Drive Itself by Imitation and Imagination
31mDeep learning for self-driving vehicle technology
58mAn introduction to Machine Learning using LEGO