Scaling reinforcement learning to large interactive environments and long-horizon robotic tasks - Naver Labs Europe