Goal exploration augmentation via pre-trained skills for sparse-reward long-horizon goal-conditioned reinforcement learning.