Is the error stated below a result of using a different version of MATLAB than the one the code in question was created with?

To preface this question, I tried to follow the steps in the video below and use an RL algorithm to simulate a walking robot using MATLAB R2020a

https://www.youtube.com/watch?time_continue=865&v=6DL5M9b2j6I&feature=emb_title

After downloading the resources in the video from his link in the description ( https://www.mathworks.com/matlabcentral/fileexchange/64227-matlab-and-simulink-robotics-arena-walking-robot?s_eid=PSM_15028 ), I tried to run the "createDDPGNetworks" file as he does as 7:07 in his video, and I got the error:

Unrecognized function or variable 'numObs'.

Error in createDDPGNetworks (line 12)

imageInputLayer([numObs 1 1],'Normalization','none','Name', 'observation')

While others have used his model and had plenty of success, I and a few others have gotten this same error. I was wondering if this is a result of him using MATLAB R2019a and me using MATLAB R2020a.

[Lanh Van Nguyen]

Hi Sammy. I have the same error as yours. Have you fixed the problem yet?

Answer 1

[Yahya Madhi]

Hello Sammy

Not sure if you have resolved the issue, but if you have not, follow the following instructions:

1. Run the script startupWalkingRobot.m
2. Open the simulink model named walkingRobotRL2D.slx
3. Run the script robotParametersRL.m
4. In the script createDDPGNetworks.m edit line 53 to critic = rlQValueRepresentation(criticNetwork,observationInfo,actionInfo,'Observation',{'observation'},'Action',{'action'},criticOptions);
5. Also in the sript named createDDPGNetworks.m edit line 85 to actor = rlDeterministicActorRepresentation(actorNetwork,observationInfo,actionInfo,'Observation',{'observation'},'Action',{'ActorTanh1'},actorOptions);
6. Thereafter open the script titled createWalkingAgent2D.m, select the appropriate Speedup Options for your system (lines 6-8). Once changed, run this script.

Hope this helps.

[IronMan]

Thanks, this was very helpful.

Answer 2

[Cam Salzberger]

Hello Sammy,

If you notice the workspace before Sebastian runs the script, it already has many variables defined. The createDDPGNetworks script is making use of some of those variables when setting up its neural networks. If you check out the createWalkingAgent2D (or 3D, I'm not sure which he was using), you can see that numObs is defined there.

-Cam

[Sammy Rossberg]

Hello Cam,

Since I aksed this question I have defined a few variables and solved a few errors, including the one above.

Unfortunately, now when I try running both createWalkingAgent2D or 3D, I get errors referring back to createDDPGNetwork. The error I get can be seen below:

Error using rlRepresentation (line 70)

rlRepresentation will be removed in a future release. Unable to automatically convert rlRepresentation to new representation

object. Use the new representation objects rlValueRepresentation, rlQValueRepresentation, rlDeterministicActorRepresentation, or

rlStochasticActorRepresentation instead.

Error in createDDPGNetworks (line 51)

critic = rlRepresentation(criticNetwork,criticOptions, ...

Error in createWalkingAgent2D (line 31)

createDDPGNetworks;

I went through the MATLAB page that explains these actors and from what I understood I should be able to replace “rlRepresentation” with something like “rlValueRepresentation” while still leaving “rlRepresentationOptions” where it appears. However, when I do that, I still get errors specifically saying:

Error using rlValueRepresentation (line 43)

Too many input arguments.

Error in createDDPGNetworks (line 51)

critic = rlValueRepresentation(criticNetwork,criticOptions, ...

Hopefully this can be easily solved and there aren’t more errors to follow. Thank you for your help so far.

- Sammy

[Cam Salzberger]

I'm sorry, but I'm not very experienced with the Reinforcement Learning Toolbox. However, if you do a search through the release notes for the removed function, it links to recommendations for how to replace the functionality. Hopefully this will be able to get you started.

-Cam