Converting and R object defined inside of a function to a Python object fails using Reticulate in RMarkdown

I want to write an R function in RMarkdown, where I declare an R object inside the function and then convert this object to a Python object that I then pass to a Python function. However, during the conversion part I get and error in py_run_string_impl(code, local, convert).

Below is a minimal working example that showcases, in principle, what I'm trying to achieve and highlights the issue that I'm having:

knitr::opts_chunk$set(echo = TRUE)

Import reticulate and define an R object:

library(reticulate)
a = 5 # Variable declared in global environment

Define an R function:

printPython = function(){
  b = 2 # Variable declared within the scope of the function
  # Attempt to translate both to python and run a python command:
  py_run_string("print(r.a)")
  py_run_string("print(r.b)")
}

Execute the function:

printPython()

This throws: Error in py_run_string_impl(code, local, convert) : RuntimeError: Evaluation error: object 'b' not found.

Re-declare 'b' in the global environment and run printPython() again:

b = 2
printPython()

The function now runs fine. If reticulate is imported inside the function the issue persists:

printPython2 = function() {
  library(reticulate)
  x=2
  py_run_string("print(r.x)", local = TRUE)
}

printPython2()

Throws the same error as before.

If I assign the declared variable to the global environment (YUCK) the function once again run fine:

printPythonGlobal = function() {
  assign("x", 2, envir = globalenv())
  py_run_string("print(r.x)", local = TRUE)
}

printPythonGlobal()

What is the proper way to do conversions of R objects declared inside of a function? How can I do this without having to resort declaring variables inside the function as global (which is generally highly discouraged)?