Srijan Kumar Pal

I'm a 1st year Ph.D student in EE at the University of Minnesota, where I work under Prof. Jiarong Hong, as a part of the Flow Field Imaging Lab.

Previously I received my Masters in Robotics from University of Minnesota and my Bachelor's degree from Jadavpur University. During my time at UMN, I have also had the opportunity to collaborate with Prof. Maria Gini and Prof. Changhyun Choi.

Email  /  Resume  /  LinkedIn  /  Github

Research Interests

I'm interested in computer vision, vision-based robot learning, and autonomous control. My research focuses on developing vision-based controllers that enable autonomous systems to perceive, navigate, and interact intelligently in dynamic, unstructured environments.

An autonomous drone controller using a PPO-based DRL framework integrated with segmentation, trained on Unreal Engine, and deployed on Jetson equipped quadcopter, for robust dynamic smoke plume tracking in unsteady winds, improving tracking efficiency (vs PID) by 15%.

A 3D path-planning framework using offline and online A* with local sensing-based replanning to navigate voxelized environments containing unknown dynamic obstacles, with interactive 3D visualizations and quantitative planning metrics.

A sparsity-aware deep super-resolution model that enforces L1 regularization on selected intermediate feature maps to balance reconstruction fidelity and perceptual quality, outperforming baseline SRCNN and VDSR variants through systematic ablation.

Developed an algorithm for autonomous drone-based image capture in optimal circular paths around the target, and using the sorted sharp image to generate camera trajectories with COLMAP, and feeding them into a modified Instant-NGP (optimized MLP and encoding) for high-fidelity 3D reconstruction.

Developed a synchronized, autonomous 4-drone swarm for adaptive multi-view imaging, dynamic repositioning, and high-fidelity NeRF-based volumetric 3D reconstructions of evolving smoke plumes at a 1-second temporal resolution.

Deployed a 4-drone swarm with DIH particle sensors in a master-worker architecture and autonomously generate waypoints to map real-time volumetric smoke particle dispersion during a large-scale prescribed forest fire, providing insights into smoke particle morphology and spatial distribution.

Source code available on John Barron's Website.