SAGE: Subgoal-Conditioned Action Generation for Latent World Model Planning
2026-07-20 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors address a problem in planning where it's hard to pick good actions when you plan far ahead because the number of possible actions grows a lot. They propose using a 'prior-conditioned planner' that predicts important intermediate goals (subgoals) before generating action sequences, helping to guide the search more effectively. By combining short- and long-term subgoals, their method improves planning for longer tasks while still doing well on shorter ones. Experiments show this approach greatly boosts success rates in benchmark tasks compared to random action proposals.
latent world modelsplanning horizonaction-conditioned dynamicssubgoalsgoal-conditioned generatorprior-conditioned plannerlong-horizon planninglatent spaceaction sequence generationmodel-based reinforcement learning
Authors
Letian Cheng, Qi Zhang, Yisen Wang
Abstract
Latent world models have emerged as a powerful planning paradigm by learning action-conditioned predictive dynamics and using them as internal simulators to imagine and evaluate candidate action sequences. However, as the planning horizon grows, performance becomes increasingly constrained by proposal quality: a fixed candidate budget must search an exponentially larger action space, making it difficult to expose the world model to high-quality candidate futures for evaluation. In this paper, we introduce a prior-conditioned planner that replaces random proposal initialization with structured guidance. At each planning stage, a goal-conditioned generator predicts the next reachable latent subgoal for a specified duration, which is then used to condition the generation of candidate action sequences. To capture semantic information across temporal scales, we use subgoals of varying durations as priors, balancing fine-grained local control with higher-level long-horizon progress. Then the frozen world model evaluates and refines these subgoal-conditioned proposals before execution. Experiments on PushT and OGBench Cube show that coupling latent subgoal decomposition with prior-conditioned action generation substantially improves long-horizon planning while preserving strong short-horizon performance. To be specific, when the target offset is $150$, it raises PushT success from $12.7\%$ to $64.7\%$ and OGBench Cube success from $26.7\%$ to $67.3\%$.