cs.CVNov 14, 2026

Toward Planning Representations That Separate Task Structure from Visual Content

Authors: Aarav Srivastava, Vishwas Sathish, Rajesh P. N. Rao

Organizations: Paul G. Allen School of Computer Science & Engineering University of Washington, Seattle, United States

Abstract

Effective planning requires representations that preserve task-relevant geometry and object identity while ignoring visual variation that does not alter the task. Existing evaluations of visual generalization often change several factors simultaneously, making the source of a failure difficult to identify. We introduce a controlled evaluation of object-goal navigation (ObjectNav), the task of reaching an instance of a named object category, in procedurally generated ProcTHOR houses. Training and evaluation houses share floor plans, object locations, goals, and start states, while visual factors are changed individually or in combination. Across model-free reinforcement learning (RL), model-based RL, and agents using frozen pretrained visual representations, our results show that appearance changes impair navigation, but their effects differ sharply by factor. Surface material changes account for most of the average change in success, whereas clutter and lighting have little effect. The magnitude of a global image statistics shift does not reliably predict performance: changing only the goal object's appearance can substantially impair an end-to-end agent despite a small image-level change. We found that a model based agent that plans over frozen features from the pretrained DINOv2 image encoder achieves the highest success (0.83) under the full shift, but remains sensitive to surface appearance. Our experiments reveal an important requirement for planning representations: they must separate invariant spatial task structure from appearance.

Figures & tables

Appendix figures & tables16 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Aug 10, 2026

Latent World Models with Monotone Planning Costs for Image-Goal Navigation

Image-goal navigation with latent world models requires not only accurate future prediction, but also a planning cost that reliably ranks candidate action sequences. We define the cost as the cosine distance between the predicted future embedding and the goal embedding, and show that poor cost ordering can mislead sampling-based planners such as Cross-Entropy Method (CEM). To address this, we propose a latent world model built on a frozen DINO-family encoder and train it with two complementary objectives. An autoregressive rollout loss reduces the gap between training and multi-step planning rollouts, while a Monotone Cost Ranking (MCR) loss directly encourages increasingly perturbed action sequences to receive higher planning costs. We also study InfoNCE-based action-contrastive training and find that temporal permutation negatives distort the latent geometry and degrade planning performance. On the GNM navigation dataset, our method outperforms Navigation World Models (NWM), DINO-WM, OmniVLA, and NoMaD, achieving state-of-the-art image-goal navigation performance while reducing orientation error by 2.7×2.7\times over the same-encoder DINO WM baseline. We also deploy the model zero-shot on a physical robot, where it follows goal-directed paths in unseen indoor and outdoor environments.
Oct 22, 2026

See What I Need, Dream What I Want: A Navigation Foundation Model on a Self-Constructed Representation Space

What visual representation should a navigation policy act on? Navigation policies built on vision foundation models largely leave representation design to the encoders' pretraining objectives. We present RepresentNav, a navigation foundation model built on a self-constructed representation space: a visual representation constructed by the policy's own objectives from vision foundation models. RepresentNav builds this space with Representation Residuals, a compact three-layer transformer that attends over features spanning the full depth of multiple complementary vision foundation models, so that the policy learns to see what it needs for navigation. In the same space it dreams what it wants: before every action a latent chain-of-thought predicts the representation of the scene a moment ahead. The dream is trained end-to-end with the policy as a joint-embedding predictive architecture (JEPA), with a next-representation prediction loss and a regularizer on its targets. RepresentNav sets a new state of the art on VLNVerse, VLN-CE R2R, EVT-Bench, and two point-goal benchmarks, and its gains carry over to a physical robot amid clutter and moving people. On VLNVerse fine-grained, RepresentNav achieves 83.80% success, outperforming the previous state of the art by 20.05 percentage points, and on EVT-Bench distracted tracking it reaches 85.5% with a single camera (+10.1). With the same encoders, the self-constructed space clearly outperforms concatenating their outputs and learns to weight their layers unevenly, and a dream in it cuts collisions by a quarter. A single dreamed latent decodes into the depth, semantics, and 3D occupancy of the scene a moment ahead, more accurately than pretrained encoders applied to the real future frame, although occupancy is never supervised, and the rollout tracks the robot's own motion over the next second: the dream emerges as a compact latent world model.
Nov 12, 2026

Learning Latent Goal Costs for Visual World-Model Planning with an Expanding Curriculum

Image-goal planning with a frozen visual world model depends on how predicted observations are scored against the goal. Mean-squared error (MSE) between embeddings from a pretrained visual encoder can provide a useful signal near a goal, yet favor states across an obstacle that are visually similar to the goal and difficult to reach. We fit a small residual adapter on top of frozen I-JEPA embeddings to improve this latent goal cost. Starting near each goal, a planner collects trajectories from progressively wider regions of initial states. The adapter learns from the order of observations along successful trajectories, with additional comparisons inferred from failed rollouts. No task-specific progress labels are used for fitting, so the procedure can be applied to other goal-reaching tasks. DotWall and PointMaze are used in this work. In paired 50-case evaluations with identical starts, goals, world models, and planner settings, success increases from 54% to 98% in DotWall and from 82% to 96% in PointMaze. On 16 unseen DotWall layout-goal cases, the adapter fitted on one fixed layout improves agreement between latent-cost ordering and shortest legal-path ordering from 71.9% to 85.8%. The results indicate that planning experience can improve a frozen visual representation's goal cost without retraining the encoder or the world model.