We propose a new way to make policy optimization more stable. - View it on GitHub