Guided Cooperation in Hierarchical Reinforcement Learning via Model-Based Rollout.

Goal-conditioned hierarchical reinforcement learning (HRL) presents a promising approach for enabling effective exploration in complex, long-horizon reinforcement learning (RL) tasks through temporal abstraction. Empirically, heightened interlevel communication and coordination can induce more stable and robust policy improvement in hierarchical systems. Yet, most existing goal-conditioned HRL algorithms have primarily focused on the subgoal discovery, neglecting interlevel cooperation. Here, we propose a novel goal-conditioned HRL framework named Guided Cooperation via Model-Based Rollout (GCMR; code is available at https://github.com/HaoranWang-TJ/GCMR_ACLG_official), aiming to bridge interlayer information synchronization and cooperation by exploiting forward dynamics. First, the GCMR mitigates the state-transition error within off-policy correction via model-based rollout, thereby enhancing sample efficiency. Second, to prevent disruption by the unseen subgoals and states, lower level Q-function gradients are constrained using a gradient penalty with a model-inferred upper bound, leading to a more stable behavioral policy conducive to effective exploration. Third, we propose a one-step rollout-based planning, using higher level critics to guide the lower level policy. Specifically, we estimate the value of future states of the lower level policy using the higher level critic function, thereby transmitting global task information downward to avoid local pitfalls. These three critical components in GCMR are expected to facilitate interlevel cooperation significantly. Experimental results demonstrate that incorporating the proposed GCMR framework with a disentangled variant of hierarchical reinforcement learning guided by landmarks (HIGL), namely, adjacency constraint and landmark-guided planning (ACLG), yields more stable and robust policy improvement compared with various baselines and significantly outperforms previous state-of-the-art (SOTA) algorithms.