Deep Reinforcement Learning for Walking Robots Using Reference Signal

Hi Everyone,

I was trying to implement what Brian suggested in this video (https://www.mathworks.com/videos/reinforcement-learning-part-4-the-walking-robot-problem-1557482052319.html) (minutes 9-12) for reference tracking. I edited the model as shown in the attached screenshots and changed the number of observations to 36 and did the training for 1000 episodes as Brian did using the same reward function he used. However, none of the trained agents could successfully walk the robot for more than one second. I was just wondering if I did something wrong or maybe the observation vector order is not correct?

Thanks a lot for the help!

[Sayak Mukherjee]

Hi

did you get the model from the video? I am also looking for the model as I am facing issue similar to yours.