Reinforcement Learning Using Action as Time series command

I am trying to solve a RL problem with two finger robot simulated in simscape. The input to the robot is a time series signal with time and gripping position (21 time points). For every action coming from RL agent has to execute this as a single time based gripping position as action to get one observation. I am not able to find a relevant example for this case. Request your support to provide guidance to solve this.

Answer 1

[Emmanouil Tzorakoleftherakis]

Let me make sure I understand the question: the RL Agent action is to close the gripper or not, but you want this action to be provided to the robot only after the time-series trajectory is executed. Is that accurate?

In that case, you could place the RL Agent block in a triggered subsystem. This woud allow you to implement the logic of when you want RL to read observations and provide an action (there is no example I can point you to for this unfortunately).