search

Unity-Technologies / ml-agents

The Unity Machine Learning Agents Toolkit (ML-Agents) is an open-source project that enables games and simulations to serve as environments for training intelligent agents using deep reinforcement learning and imitation learning.
https://unity.com/products/machine-learning-agents
Other
16.98k stars 4.14k forks source link

Q-learning vs Evolution algorithm #290

Closed svLimones closed 6 years ago

svLimones commented 6 years ago

Greetings I'm trying to train a navigation agent. A simple field algorithm. But I can not understand why training is so ineffective. The genetic algorithm allows to find a good agent for 1000 steps. Q-learning does not reach such a result and for 200,000 steps. Tried various parameters of training. 1) Q-learning is real less effective for such an environment? 2) Is it possible to integrate an evolutionary algorithm into the training process? 3) Is it possible to teach a previously trained agent? thanks for the interesting framework

awjuliani commented 6 years ago

Hi @svLimones,

Thanks for reaching out. The relative performance of these kinds of algorithms can vary quite a bit based on the implementation and the problem you are trying to solve. We are actively examining whether to include DQN variants or ES algorithms. If you have a particular algorithm you'd like to see implemented, please feel free to create a new issue an tag it with request.

lock[bot] commented 4 years ago

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.