BATE

A bottleneck-adaptive targeted exploration method within the AtomicRL framework that dynamically allocates sampling weights and exploration incentives according to the degree of mastery of atomic subtasks, realizing data-driven adaptive curriculum learning.