Encountering problems in creating a Simulink interactive time-sequenced reinforcement learning environment.

I want to set up an online learning environment for PPO in Simulink, and the status input is 2*100 time series data, and I would like to ask how to implement it. observationInfo = rlNumericSpec([2 100]); When I use this code, I get an error.

错误使用 rl.internal.validate.mapFunctionObservationInput (第 50 行)

Model input sizes must match the dimensions specified in the corresponding observation and action info specifications.

出错 rlDiscreteCategoricalActor (第 86 行)

model = rl.internal.validate.mapFunctionObservationInput(model,...

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

出错 rl_demo (第 80 行)

actor = rlDiscreteCategoricalActor(actorNetwork,observationInfo,actionInfo);

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Answer 1

[Kautuk Raj]

I can see that you are facing the error "Model input sizes must match the dimensions specified in the corresponding observation and action info specifications".

I had also faced the same error during one of my workflows.

I updated my MATLAB version to R2024a (or beyond) and this fixed it for me.

Suggesting you to try the same!

[浩文]

My matlab version is 2024b, and the observation space is changed to [2 1] network output layer: sequenceInputLayer(2). It's not working, please show me the code, let me refer to check where my problem is, I would be very grateful.