WoRV Research

Research papers and projects from the WoRV team.

About Us

WoRV (World Models for Robotics and Vehicle Control) Team has been pioneering VLA-based RFM (Robotic Foundation Models) for agricultural, logistics, defense, construction, and manufacturing domains since 2021.

We go beyond "fancy lab research." We operate in the real world with a fleet of hundreds of robots replacing human labor in industrial sites. This allows us to collect robust training data, establish new research hypotheses daily, and immediately evaluate and realign our models through field deployment and business feedback.

We offer the optimal environment to "pre-train" for the upcoming era of Physical AI.

Research Highlights

VLA Evaluation Harness From Data to Decisions Workshop @ ICRA 2026

A unified evaluation harness for Vision-Language-Action models — accepted at the ICRA 2026 Workshop on From Data to Decisions.

View on GitHub arXiv View Leaderboard

CostNav ScaleBot Workshop @ CVPR 2026

Economic-Aware Navigation Benchmark for Autonomous Delivery Robots

The first navigation benchmark that evaluates robots by profit per run. CostNav bridges the gap between impressive research demos and sustainable businesses by modeling the complete economic lifecycle of delivery robots.

View Project Page arXiv

D2E ICLR 2026

Scaling Vision-Action Pretraining on Desktop Data for Transfer to Embodied AI

Leveraging large-scale desktop data to pre-train vision-action models, demonstrating effective transfer to embodied robotic tasks.

View Project Page arXiv

WorldCam VideoWorldModel Workshop @ CVPR 2026

Interactive Autoregressive 3D Gaming Worlds with Camera Pose as a Unifying Geometric Representation

A generative world model that builds interactive 3D worlds from gameplay data, using camera pose as a unifying geometric representation. A collaboration between KAIST CVLab, Adobe Research, and WoRV @ Maum.ai.

View Project Page arXiv

SketchDrive (CANVAS) ICRA 2025

Commonsense-Aware Navigation System for Intuitive Human-Robot Interaction

A novel navigation system that interprets user sketches to guide robots, bridging the gap between human intuition and robot control.

View Project Page arXiv

We are Hiring!

Join a team of top-tier talents tackling the hardest problems in Physical AI.

We promise industry-leading compensation, alternative military service options, and the most reliable colleagues.