AWS DeepRacer
Developer Guide

This is prerelease documentation for a service in preview release. It is subject to change.

Choose a Reinforcement Learning Framework and Algorithm

Using a machine learning framework to train any machine learning model is useful. It allows you to focus on things specific to your system while leveraging tested general-purpose implementations of underlying algorithms. You should use one whenever possible.

To train your reinforcement learning model for AWS DeepRacer, you can choose one of the following currently supported deep-learning frameworks and algorithms for you to train a reinforcement learning model for AWS DeepRacer.

Reinforcement learning frameworks and algorithms supported by AWS DeepRacer vehicles

Frameworks Algorithm Comment
TensorFlow PPO The proximal policy optimization (PPO) algorithm achieves policy gradient convergence by introducing a penalty to the objective function to represent the constraint on the policy gradient.