See https://github.com/HumanCompatibleAI/imitation/pull/657 - replace hardcoded paper results with a CSV of results that we can compare and update Also, include instructions for updating the CSV, and a comparison between paper results and current benchmark results
