Search papers, labs, and topics across Lattice.
This paper introduces SAGE, a novel planning approach that enhances latent world models by integrating subgoal-conditioned action generation to improve long-horizon planning. By replacing random proposal initialization with structured guidance from a goal-conditioned generator, the method effectively balances local control and long-term progress through the use of subgoals of varying durations. Experimental results demonstrate significant improvements in planning success rates on both PushT and OGBench Cube tasks, highlighting the method's efficacy in navigating complex action spaces.
Subgoal-conditioned action generation boosts long-horizon planning success rates from 12.7% to 64.7% in complex environments.
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 $67.3\%$.
