Accelerative Synthetic Data Generation
Intelligent diffusion-based filtering to detect and remove inauthentic synthetic videos 9× faster with early exit diffusion pipelines. Achieves 75% compute savings and 6× faster video generation.
Projects
A showcase of my work in robotics, generative AI, and intelligent systems. Each project represents a step toward making robots more capable and intelligent.
Generative AI
Diffusion models, synthetic data generation, and foundation models
Ortho-Fuse: Orthomosaic Generation for Crop Health
Uses intermediate optical flow estimation for synthetic frame generation in orthomosaic creation from sparse aerial images. Enables high-quality crop health analysis with reduced image overlap (50% vs. traditional 70-80%).
Generative AI + Robotics
Combining foundation models with robotic systems for intelligent behavior
World Foundation Models + VLA Integration
Augmented real-world video datasets for VLA training, addressing data scarcity with minimal trajectories. Enables efficient policy training maximizing task performance and scalable robotic learning.
Physics-based Digital Twin for Wildlife Monitoring
Photorealistic NVIDIA Omniverse simulation with generative animal behaviors, herd dynamics, and drone responses. Reduces field deployment costs and accelerates algorithm prototyping.
SmartWilds: Multimodal Wildlife Dataset
Generated synthetic multimodal dataset for wildlife behaviors using world models, supporting privacy-constrained ecological AI. Boosts robust classification from scarce data with drone and camera-trap integration.
Robotics
Autonomous systems, multi-agent coordination, and robot control
Autonomous Vineyard Rover for Data Collection
Deployed rover with segmentation neural network and depth sensing for lane centering on uneven terrain. Achieves robust navigation with 40° slip compensation in real field tests.
Multi-Agent RL for Optimal Field Scouting
CNN-integrated reinforcement learning framework for heterogeneous agents in agricultural scouting. 60% reduction in scouting needs, 80% accuracy, 4.8× labor cost savings, 36% farmer profit boost.
Decentralized Autonomous Intersection Management
Level 3.5 motion and behavior planning using road-marking intent detection and graph-theoretic coordination. Communication-free, deadlock-free navigation across 255 scenarios.
Priority-Balanced Patrol Planning
Distributed online patrol-planning algorithm with scalable trajectories balancing priority and non-priority site coverage. Finite-time visit guarantees with sim-to-real validation.
Swarm Synergy: Communication-Free Formations
Autonomous community formations for multi-robot swarms without communication. Validated in simulations, experiments, and real-world lab tests for reliable deployment.
Heterogeneous UAV Swarm for Crop Monitoring
Multi-agent reinforcement learning-based coordination for UAVs in detailed crop health assessment. Scalable multi-agent deployment without reliance on synthetic data.