Safe Reinforcement Learning through Transferable Instinct Networks - View it on GitHub