Allow passing a params dictionary into env.step()

[JohnBoik]

When creating a multi-agent system, it can be useful to pass objects (observations, states, B matrices, etc.) between different environments (if each agent has an environment). However, probably because of jit, one cannot effectively alter the B matrix of an environment using something like:

env.params[“B”][2] = new_B_matrix

Doing so might not change env.params[“B”] as it is used in computations. A possible solution is to pass in an altered env.params[“B”] to env.step(). In that case, the passed version would be used in computations instead of self.params[“B”].

[JohnBoik]

I expand this issue here to also include the ability to pass A and B matrices into agent functions. This would allow for manual updates to those matrices within a simulation step. In essence, calls to self.A and self.B internal to pymdp would need to be replaced with the A and B parameters that are passed into functions.