A striking example of how not to present data. If the Cognizant AI team is here: please can you fix it in the next version of the paper?
As someone else points, the data is the worrying aspect, as it points towards state-of-the-art models not being able of making more than 0 consecutive steps without errors.