World Foundation Models
Augmenting video data and synthesizing trajectories for sample-efficient policy learning.
Available · Summer 2026 · Industry research
Building world models that move atoms.
I'm Rugved Katole — PhD candidate at The Ohio State University researching vision-language-action systems and world foundation models for general-purpose robotics. Focused on closing the data gap between simulation and reality.
- 4+ Years in robotics
- 9 Peer-reviewed publications
- $100K Research grant led
- 5.9× Inference speedup
/ about
From mechanical hands to general-purpose minds.
Trained as a mechanical engineer at BITS Pilani and now a PhD candidate in Computer Science at The Ohio State University, I work at the intersection of foundation models and embodied AI. Previously at IIT Bombay's TIH and ARMS Lab, leading deployments in autonomous navigation and multi-agent planning.
I care about systems that work outside the lab — robots that handle 40° slips on vineyard slopes, UAV swarms that map fields without choreography, and policies trained from scarce, real-world data.
Vision-Language-Action
Closing the loop between perception, language, and embodied control across manipulation tasks.
Sim-to-Real Robotics
Photoreal Omniverse digital twins, ROS 2 deployments, and field-tested multi-agent systems.
Research that ships into the real world.
Six projects spanning foundation models, multi-agent autonomy, and embodied AI. Each connects to a deployed system, a published paper, or both.
01
World Foundation Models × VLA
Augment real-world video datasets for vision-language-action training. Synthesize counterfactual trajectories to make robotic policies sample-efficient.
data min trajectories
gain few-shot
02
Accelerative Synthetic Data
Diffusion-based filtering and early-exit pipelines that detect inauthentic synthetic videos 9× faster — 75% compute saved on generation.
9× faster filter
75% compute saved
03
Wildlife Digital Twin
Photoreal NVIDIA Omniverse simulation of generative animal behaviors and herd dynamics. Drone algorithms validated before field deployment.
∞ sim hours
↓ field cost
04
Decentralized AV Intersection
Communication-free, deadlock-free intersection coordination using graph theory and road-marking intent detection across 255 scenarios.
255 scenarios
0 deadlocks
05
Multi-Agent UAV Swarm
CNN + multi-agent RL for heterogeneous UAV crop scouting. Cuts scouting need 60%, labor cost 4.8×, lifts farmer profit 36%.
60% less scouting
36% profit ↑
06
Priority Patrol Planning
Distributed online patrol with finite-time visit guarantees, balancing priority and non-priority site coverage. Sim-to-real validated.
O(n) scalable
✓ sim2real
Recent peer-reviewed work.
Selected publications across world models, edge computing for conservation, and multi-agent autonomy. Google Scholar →
- 2025
Ortho-Fuse: Orthomosaic Generation for Sparse High-Resolution Crop Health Datasets
ICPP — International Conference on Parallel Processing - 2025
Edge-Native, Behavior-Adaptive Drone System for Wildlife Monitoring
ACM/IEEE Symposium on Edge Computing (SEC) - 2025
SmartWilds: Multimodal Wildlife Monitoring Dataset
arXiv:2509.18894 - 2023
A Low-Cost Framework for Decentralized Autonomous Intersection Management
arXiv:2311.17681
/ collaborate
Let's build the next embodied agent.
I'm open to industry research roles, collaborations on world models & VLA, speaking, and consulting. Best reached over email or LinkedIn.