Depth-Regularized JEPA World Models Learn More Transferable Representations from Real Outdoor Robot Data

• URL: http://arxiv.org/abs/2607.16314v1
• Date: Wed, 15 Jul 2026 00:58:45 GMT
• Title: Depth-Regularized JEPA World Models Learn More Transferable Representations from Real Outdoor Robot Data
• Abstract summary: We introduce depth as a geometric prior during training in learning more robust latent dynamics directly from robot video data.<n>We evaluate with frozen-representation visual odometry probes, predictor-based surprise detection, and multi-step latent rollout fidelity.<n>Results show that a lightweight training-time geometric prior makes a compact JEPA world model more useful and more transferable on real outdoor data.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: World models, especially based on JEPA architectures, have been shown to learn robust dynamics of various environments. However, learning from visually complex real-world data remains a challenge, especially in unpredictable outdoor environments. We introduce depth as a geometric prior during training in learning more robust latent dynamics directly from robot video data and handling visual complexity. This combines depth supervision with an isotropy-inducing latent regularizer (SIGReg), maximizing task-agnostic latent diversity while constraining how that diversity is organized, with the combined objective targeting the highest-entropy representation consistent with scene geometry. To satisfy this greater complexity without increasing inference time, we also add training-only overparameterization. Training an 18M-parameter model on video from a real agricultural robot, we evaluate with frozen-representation visual odometry probes, predictor-based surprise detection, and multi-step latent rollout fidelity. Compared to the baseline LeWM, our method lowers visual odometry probe error by 33%, substantially increases surprise-score separation both in-domain and on the out-of-domain TartanGround benchmark, and improves multi-step rollout fidelity under domain shift, with gains that grow with rollout horizon. Notably, we also see improvements in surprise-score separation on physics understanding that is not directly tied to 3D geometry, such as lighting and shadows. These results show that a lightweight training-time geometric prior makes a compact JEPA world model more useful and more transferable on real outdoor data with strong underlying representations, without adding inference overhead. Our work suggests that depth as a physically grounded prior can enhance world model generalization on a variety of tasks.

Related papers

• Generative Texture Diversification of 3D Pedestrians for Robust Autonomous Driving Perception [0.0]
  In this paper, we propose a method for scaling variability in 3D pedestrian assets for synthetic scene generation.<n>We generate multiple distinct pedestrian instances by synthesizing diverse facial textures and identity-level appearance variations.<n>Using the assets, we construct synthetic datasets and study the impact of mixing real and synthetic data for RGB-based object detection.
  arXiv  Detail & Related papers  (2026-05-13T16:35:50Z)
• Mirage2Matter: A Physically Grounded Gaussian World Model from Video [87.9732484393686]
  We present Simulate Anything, a graphics-driven world modeling and simulation framework.<n>Our approach reconstructs real-world environments into a photorealistic scene representation using 3D Gaussian Splatting (3DGS)<n>We then leverage generative models to recover a physically realistic representation and integrate it into a simulation environment via a precision calibration target.
  arXiv  Detail & Related papers  (2026-01-24T07:43:57Z)
• UniSH: Unifying Scene and Human Reconstruction in a Feed-Forward Pass [83.7071371474926]
  UniSH is a unified, feed-forward framework for joint metric-scale 3D scene and human reconstruction.<n>Our framework bridges strong, disparate priors from scene reconstruction and HMR.<n>Our model achieves state-of-the-art performance on human-centric scene reconstruction.
  arXiv  Detail & Related papers  (2026-01-03T16:06:27Z)
• MoRE: 3D Visual Geometry Reconstruction Meets Mixture-of-Experts [50.37005070020306]
  MoRE is a dense 3D visual foundation model based on a Mixture-of-Experts (MoE) architecture.<n>MoRE incorporates a confidence-based depth refinement module that stabilizes and refines geometric estimation.<n>It integrates dense semantic features with globally aligned 3D backbone representations for high-fidelity surface normal prediction.
  arXiv  Detail & Related papers  (2025-10-31T06:54:27Z)
• Clone Deterministic 3D Worlds with Geometrically-Regularized World Models [16.494281967592745]
  World models are essential for enabling agents to think, plan, and reason effectively in complex, dynamic settings.<n>Despite rapid progress, current world models remain brittle and degrade over long horizons.<n>We propose Geometrically-Regularized World Models (GRWM), which enforces that consecutive points along a natural sensory trajectory remain close in latent representation space.
  arXiv  Detail & Related papers  (2025-10-30T17:56:43Z)
• R2RGEN: Real-to-Real 3D Data Generation for Spatially Generalized Manipulation [74.41728218960465]
  We propose a real-to-real 3D data generation framework (R2RGen) that directly augments the pointcloud observation-action pairs to generate real-world data.<n>R2RGen substantially enhances data efficiency on extensive experiments and demonstrates strong potential for scaling and application on mobile manipulation.
  arXiv  Detail & Related papers  (2025-10-09T17:55:44Z)
• RLGF: Reinforcement Learning with Geometric Feedback for Autonomous Driving Video Generation [75.61028930882144]
  We identify and quantify this critical issue, demonstrating a significant performance gap in 3D object detection when using synthetic versus real data.<n>We introduce Reinforcement Learning with Geometric Feedback (RLGF), RLGF uniquely refines video diffusion models by incorporating rewards from specialized latent-space AD perception models.<n> RLGF substantially reduces geometric errors (e.g., VP error by 21%, Depth error by 57%) and dramatically improves 3D object detection mAP by 12.7%, narrowing the gap to real-data performance.
  arXiv  Detail & Related papers  (2025-09-20T02:23:36Z)
• Enhancing Generalizability of Representation Learning for Data-Efficient 3D Scene Understanding [50.448520056844885]
  We propose a generative Bayesian network to produce diverse synthetic scenes with real-world patterns. A series of experiments robustly display our method's consistent superiority over existing state-of-the-art pre-training approaches.
  arXiv  Detail & Related papers  (2024-06-17T07:43:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.