V-MIN: Efficient Reinforcement Learning through Demonstrations and Relaxed Reward Demands