Initializing

BHARATH VEDANTHA DESIKAN

Robotics Research Engineer

Scroll

Building smarter
autonomous systems

I'm Bharath Vedantha Desikan, a robotics and machine learning researcher at Arizona State University's DREAMS Lab, where I develop algorithms that enable autonomous platforms to efficiently explore and reconstruct spatial environments.

My work sits at the intersection of probabilistic modeling, information theory, and robotic decision-making. I build systems that reason about uncertainty to make smarter sampling decisions, whether that means mapping aquatic temperature fields with autonomous surface vehicles or reconstructing 3D objects with aerial drones.

I focus on turning principled mathematical frameworks into real, deployable robotic systems -- from Gaussian Process regression and information-theoretic planning to 3D Gaussian Splatting for active reconstruction.

Adaptive Sampling

Information-theoretic path planning with Gaussian Processes under positional uncertainty. Seven distinct planners compared across stationary and nonstationary kernels.

3D Reconstruction

Active 3D Gaussian Splatting with next-best-view planning. Autonomous drones that reconstruct objects in real-time using depth-variance guided exploration.

Deployed Systems

Full ROS 2 / PX4 simulation pipelines with Docker orchestration, automated trial management, and statistical analysis frameworks.

Research output

icra-2026-aquatic-field-reconstruction.pdf
Published
IEEE ICRA 2026

Low Cost ASV for High-Resolution Spatio-Temporal Aquatic Field Reconstruction via Dynamic Kernels

Rodney Staggers Jr*, Bharath Vedantha Desikan*, Jnaneshwar Das — *equal contribution

case-2026-active-3dgs.pdf
Under Review
IEEE CASE 2026

Depth-Variance Guided Next-Best-View Planning for Active 3D Gaussian Splatting Reconstruction

Bharath Vedantha Desikan, Jnaneshwar Das

Paper Preview
Download

Loading paper preview...

Current work

Aquatic Field Estimation

GP-based adaptive sampling with 7 distinct planners under positional uncertainty. Autonomous surface vehicles reconstruct aquatic temperature fields using information-theoretic planning with stationary and nonstationary Gibbs kernels.

ROS 2 GPyTorch PX4 Gazebo Docker
Survey Data GitHub

Active 3D Reconstruction

Autonomous object reconstruction using depth-variance guided next-best-view planning with 3D Gaussian Splatting. Real-time incremental reconstruction with live splat optimization and voxel-based coverage tracking.

3DGS gsplat PX4 Gazebo CUDA

Tech stack

ROS2 Python Gaussian Processes PX4 Autopilot Docker 3D Gaussian Splatting PyTorch SLAM Gazebo Informative Path Planning Kernel Methods Machine Learning Computer Vision GPyTorch C++ Linux CUDA OpenCV Git FastAPI MATLAB Bash ROS1 scikit-learn LaTeX Probability Theory Stochastic Processes Linear Algebra Autonomous Mobile Robots Uncertainty-aware Planning
All Robotics Math & ML Languages Infrastructure

Research Focus

Gaussian Processes Information-Theoretic Planning Nonstationary Kernels 3D Gaussian Splatting Positional Uncertainty

Get in touch