Leveraging an adaptive skill distribution based on local entropy optimization can significantly improve exploration efficiency in goal-conditioned reinforcement learning tasks with sparse rewards and long horizons.
