Difference between using a rlContinuo​usGaussian​Actor and a rlContinuo​usDetermin​isticActor with a Gaussian Explorationmodel

Hi,

Someone please explain the difference between using a rlContinuousGaussianActor and using a rlContinuousDeterministicActor with a Gaussian Explorationmodel (namely the GaussianActionNoise) in reinforcement learning with e.g. a rlTD3agent.

With possibly using the rlContinuousGaussianActor using the Gaussian Explorationmodel would not be impossible: What is the point/ use case of this combination?

Thank you.

[Jonas Woeste]

Despite rlContinuousGaussianActor not being applicable to rlTD3agent, some explanation about similarities of the GaussianActionNoise and the rlContinuousGaussianActor would be useful.

Answer 1

[Manas]

Hi Jonas Woeste,

I understand that you wish to know the difference between "rlContinuousGaussianActor" and "rlContinuousDeterministicActor" with a Gaussian exploration model.

The choice between these two approaches depends on the specific problem and the trade-off between exploration and exploitation.

“rlContinuousGaussainActor” is suitable when exploration is crucial and stochastic actions are desired. “rlContinuousDeterministicActor” is suitable for providing a stable policy with controlled exploration. It strikes a balance between exploration and exploitation. To introduce exploration, a Gaussian exploration model like “GaussianActiveNoise” is used. The exploration model adds Gaussian noise to the actor's output action, making it slightly perturbed and allowing exploration.

In summary, the choice of the combination depends on the nature of the problem and the desired behaviour of the agent you are seeking.

Hope this helps!