🤖 AI Summary
This work challenges the common assumption that poor long-horizon planning stems primarily from inaccurate world model predictions, demonstrating instead that suboptimal planner objective functions are often the true bottleneck. Analyzing latent-variable world models in a TwoRoom environment, the authors show that these models accurately encode distant future states, yet conventional squared Euclidean distance objectives in latent space impede effective planning. By simply replacing the planning objective—without retraining the model—they boost task success rates from 26.0% to 98.0% under a 100-step horizon, and achieve 92.0% success using only one-third of the original computational budget. Through cross-entropy method (CEM) planning, latent-space geometric probing, and multi-objective comparisons, the study reveals that maximal planning performance does not require maximal prediction accuracy; rather, it hinges on aligning the objective function with the underlying task structure.
📝 Abstract
Latent world models are judged by how well they predict, so when planning fails at long horizons the natural reading is that the predictor degrades. On a reproduction of LeWorldModel on TwoRoom we show the binding constraint is the planner's objective instead.
The predictor is not the limit: its imagined state seventy-five environment steps ahead is still only 0.189 as wrong as assuming the world froze, while the planner never imagines beyond twenty-five. The objective is. Cross-entropy-method planning minimises squared latent distance, which tracks true distance at r = 0.426, saturates by about eighty arena units and decreases beyond a hundred and twenty, so moving away from the goal can lower the cost. The information is present throughout: a ridge probe recovers position from the frozen embedding at R^2 0.9922.
The pathology is the method's, not one reimplementation's. It is present in the authors' released weights, and across four checkpoints long-horizon success rank-orders exactly with metric quality and inversely with prediction accuracy.
Replacing only the objective, with nothing retrained and no GPU, lifts goals reached at offset 100 from 26.0% to 98.0%, equals the 98.0% at offset 25, and reaches 92.0% under a third of the budget: planning stops depending on the horizon. The best cost is not the most accurate. A head learned from frame separation alone predicts spatial distance worse than a position probe (r = 0.819 against 0.9897) yet plans better, charging 24% more to cross the environment's dividing wall where squared latent distance charges 4% less. It has learned reachability, not proximity.