Curiosity-driven Exploration in Sparse-reward Multi-agent Reinforcement Learning