Thanks for your advice. I agree with you that when possible, writing these integration tests in a language such as Python will be much quicker than using C++/gtest.
In the context of our (research) project, we want to create a (ideally language agnostic) framework for specifying tests with pre-conditions, invariants and post-conditions. These conditions would be verified using both intrinsic data from the robot and data from an external sensing system such as a laser scanner, rgbd-camera or mocap. It’s likely that neither pytest or gtest will be of much use, as we are interested in contributing towards better tools within ROS for constructing these kind of tests.