Neural Network Dynamics for Model-Based Deep Reinforcement Learniing with Model-Free Fine-Tuning

Goal

1. 怎样在model-based reinforcement learning中使用neural-network创建system dynamics
2. 怎样使用model-based reinforcement learning来加速model-free reinforcement learning

Related Work

The most efficient model-based algorithms have used relatively simple function approximators, such as Gaussian processes, time-varying linear models, and mixtures of Gaussians.

Contribution

1. demonstrate effective model-based reinforcement learning with neural network models for several contact-rich simulated locomotion tasks from standard deep reinforcement learning benchmarks
2. evaluate a number of design decisions for neural network dynamics model learning
3. show how a model-based learner can be used to initialized a model-free learner to achieve high rewards while drastically reducing sample complexity

The learned model-based controller provides good rollouts, which enable supervised initialization of a policy that can then be fine-tuned with model-free algorithms, such as policy gradients.

Code（这篇文章的github repository的结构还可以）