Data Analytics
Self Driving Cars Using DDQN
A Double Deep Q-Network agent that learns to drive a car around custom tracks, with a simulator to draw tracks, train, save and replay models.
- Type
- End-of-study project, self-driving car specialization
- Role
- Machine learning engineer
- Period
- Sept 2021 to May 2022
- Status
- Completed
DDQN
algorithm, implemented in TensorFlow
247
days from start to final report
Live
training view with scores and learning rate
The idea
Teach a car to drive from raw sensor input with no labelled data, and make the training process visible enough to tune and explain.
What I built
A TensorFlow implementation of Double DQN with an experience-replay buffer and a Pyglet simulator. Tracks and goals are drawn in the tool, models can be saved and loaded, and learning curves and scores are shown live.
The result
Delivered as the final project of the specialization, with a written article and a project report.
Key features
- Track and goal drawing tool
- Save and load trained models
- Live learning-rate and score display
- Manual driving mode for comparison
Screens
Track1 / 3
Stack and tools
- TensorFlow
- Deep Learning
- DDQN
- Game Design
- Pyglet
- Notebook
- Python
Code Samples
DDQN Model
LoadingNeed something like this built?
Book a free 30-minute call to talk through your project. After the discovery call, once the scope is clear, you get a fixed written quote in MAD.
