Problem
Research questions and friction points this paper is trying to address.
image-goal navigation
latent world models
planning costs
cost ranking
action sequences
Latent World Models with Monotone Planning Costs for Image-Goal Navigation
🤖 AI Summary
This work addresses the challenge in visual goal-directed navigation where existing latent world models often mislead planners due to poorly ordered cost functions. The authors propose a novel latent world model integrating a frozen DINO encoder, autoregressive rollout prediction, and a new Monotonic Cost Ranking (MCR) loss that directly optimizes the monotonicity of costs along perturbed action sequences. They further reveal that InfoNCE-based temporal negative sampling distorts the geometry of the latent space, degrading planning performance. Evaluated on the GNM dataset, the method substantially outperforms NWM, DINO-WM, OmniVLA, and NoMaD, reducing directional error by 2.7× compared to baseline models using the same encoder, and demonstrates successful zero-shot navigation on real robots in both indoor and outdoor environments.
📝 Abstract
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 $2.7\times$ 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.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Latent World Model
Monotone Cost Ranking
Image-Goal Navigation
DINO Encoder
Cross-Entropy Method
🔎 Similar Papers
No similar papers found.
