is generatePolicyFunction still not supported for agents with RNN

is generatePolicyFunction still not supported for agents with RNN for C++ hardware deployment and is there a solution?

Answer 1

[Dhruv]

The ‘generatePolicyFunction’ is not supported for agents with RNN (Recurrent Neural Network) when it comes to C++ hardware deployment. For detailed information, you can refer to the documentation available: https://in.mathworks.com/help/releases/R2023a/reinforcement-learning/ug/deploy-trained-reinforcement-learning-agents.html

The documentation explicitly mentions in paragraph two that " Code generation is not supported for continuous actions PG, AC, PPO, and SAC agents using a recurrent neural network (RNN)."

If your agent relies on an RNN architecture, it is important to note that it may not be directly deployable using 'generatePolicyFunction' for C++ hardware deployment.

In such cases, a potential workaround could involve switching to a feedforward neural network, even though RNNs are generally better suited for sequential data. However, this may be a feasible option if your task is relatively straightforward.