Paper Library

Follow-on V-JEPA line; useful for dense feature representations.

Short 2026 variant focused on regularization for world models.

Follow-up analysis of when LeJEPA learns world models.

The "LeWorldModel / LeWM" style paper you called out.

Causal/object-level extension of JEPA-style world models.

Probabilistic world-model variant of JEPA.

Small but directly relevant planning paper.

SSMG-Nav: semantic skeleton memory graph for lifelong ObjectNav

Zilong Deng, Federico Tombari, Marc Pollefeys, Johanna Wald, Daniel Barath

OVI-MAP: open-vocabulary instance-semantic mapping

Jiahua Pei, Yi Liu, Guoping Pan, Yuanhao Jiang, Houde Liu, Xueqian Wang

OVAL: open-vocabulary augmented memory model for lifelong ObjectNav

Real-time VLM indoor navigation targeting low-cost local hardware and small robots.

The LeJEPA basis that LeWorldModel builds around.

Analysis of physical planning with JEPA world models.

Explicit JEPA world-model formulation.

Direct bridge from video JEPA to planning/world-model behavior.

Fujing Xie, Sören Schwertfeger, Hermann Blum

osmAG-LLM: moved/unmapped objects in open-vocab ObjectNav

Utkarsh Bajpai, Julius Rückin, Cyrill Stachniss, Marija Popović

Uncertainty-informed active perception for OVON

ASCENT: floor-aware zero-shot object-goal navigation

Xun Huang, Shijia Zhao, Yunxiang Wang, Xin Lu, Wanfa Zhang, Rongsheng Qu, Weixin Li, Yunhong Wang, Chenglu Wen

MSGNav: multimodal 3D scene graph for navigation

Sebastián Barbas Laina, Simon Boche, Sotiris Papatheodorou, Simon Schaefer, Jaehyung Jung, Helen Oleynikova, Stefan Leutenegger

FindAnything: open-vocab object-centric mapping

DIV-Nav: spatial relationship queries with semantic maps

C-NAV: continual object navigation benchmark/framework

Semantic maps plus LLM reasoning for ObjectNav with moved/unmapped objects; relevant if small car needs semantic target navigation.

VLM reasoning distilled into attention maps for socially aware robot navigation; real Husky validation.

Real robot VLM navigation with scene graph/image memory and long indoor/outdoor runs.

Traversability-aware topological-to-local navigation, ICRA 2025, directly relevant to field navigation stacks.

Cross-modal UGV/UAV relative localization and SLAM, relevant to ground vehicle relocalization robustness.

Unified RL navigation framework with sim-to-real on multiple robots including wheeled ground vehicle.

End-to-end skid-steer robot navigation and safety control reference; practical small-vehicle system angle.

LLM-generated LTL task planning over semantic occupancy maps, evaluated on real mobile robot.

Hybrid graph search plus DRL local planning for mobile robot navigation in unknown environments.

LiDAR place recognition / localization across ground and aerial views in forest and urban scenes.

Forest LiDAR place recognition and 6-DoF metric localization for field robots.

Low-height robot viewpoint mismatch and real deployment; relevant to small chassis camera height.

UGV navigation with UAV terrain scouting and obstacle-aware replanning; hardware demo on GEM e6 ground vehicle.

Lightweight onboard visual navigation hazard avoidance on Jetson/mobile robot; engineering-relevant.

LiDAR SLAM robustness under dynamic objects and unstructured environments for mobile robot navigation.

Canonical video JEPA paper.

Hang Yin, Xiuwei Xu, Zhenyu Wu, Jie Zhou, Jiwen Lu

SG-Nav: online 3D scene graph prompting

Vishnu Sashank Dorbala, Bhrij Patel, Amrit Singh Bedi, Dinesh Manocha

TAS: navigation with non-stationary targets

Yuxuan Kuang, Hai Lin, Meng Jiang

OpenFMNav: open-set zero-shot ObjectNav

Joseph Rowell, Lintong Zhang, Maurice Fallon

LiSTA: object-level change detection with multi-mission SLAM

GOAT-Bench: multimodal lifelong navigation benchmark

Online 3D scene graph plus LLM navigation; relevant to semantic map navigation for small robots.

VLM-GroNav on Husky-like outdoor navigation with physically grounded terrain traversability.

Online lifelong VLM social navigation with real-world deployment relevance.

Systems-first open-vocabulary mobile manipulation/navigation in real homes; older than two-year window by a few months but highly useful.

Canonical image JEPA paper.

Conceptual source for JEPA, energy-based world models, and autonomous intelligence framing.

POCD: probabilistic object-level change detection

Detecting Invalid Map Merges in Lifelong SLAM

Lifelong LERF: semantic inventory monitoring

SceneVGGT: online 3D semantic SLAM for navigation

LEGO-SLAM: language-embedded Gaussian SLAM

ObjectReact: object-relative control/topometric scene graph

Open Scene Graphs / OSG Navigator

TrajRAG: retrieves geometric-semantic experience for ObjectNav

ConsistNav: persistent candidate memory / semantic executive