5. Discrete control and continuous control

Discrete VS Continuous Control

Discrete

Continuous

DQN One action, one dimension , Cannot be used for continuous control

Policy Network One action, one dimension , Cannot be used for continuous control

Must use DQN Do continuous control , It is necessary to discretize the continuous space


Better Approaches to Continuous Control

Deterministic policy network

updating Value Network by TD

Updating Policy Network by DPG

improvement:Using Target Networks

How to improve

Stochastic Policy for Continuous Control

Policy Network

Univariate Normal Distribution

Multivariate Normal Distribution

Function Approximation

Training Policy Network

Auxiliary Network

Policy Gradient Methods