[Feature Request] Concatenate the additional state as the input in network architecture

[Liuzy0908]

🚀 Feature

In autonomous driving task, there is often such a problem: the single frame image does not have the ability to perceive speed, so it is very necessary that the additional vehicle speed as the input state.

Therefore, in reinforcement learning network, observation is first reduced through CNN, and then the vehicle speed is concatenated in the last MLP layers. As shown in the follow figure.

So, does SB3 already implement this function? Or how do I implement this concatenation in SB3?

Thanks!

Motivation

In autonomous driving task, there is often such a problem: the single frame image does not have the ability to perceive speed, so it is very necessary that the additional vehicle speed as the input state.

Pitch

No response

Alternatives

No response

Additional context

No response

Checklist

• [X] I have checked that there is no similar issue in the repo

[araffin]

For autonomous driving, you can take a look at part2 of: https://www.youtube.com/watch?v=ngK33h00iBE&list=PL42jkf1t1F7dFXE7f0VTeFLhW0ZEQ4XJV

And also look in the documentation about multi input.

Edited version: https://www.youtube.com/watch?v=jXVUig0muFI

[Liuzy0908]

Thanks. 👍