Sacchin Sundar

I am a Master's student in Robotics at the University of Michigan and a member of the Field Robotics Group (FRoG), advised by Prof. Katie Skinner. My research focuses on learning representations that support perception, planning, and control, with an emphasis on neural scene representations as substrates for action. I am particularly interested in how physically grounded and structured representations can bridge low-level sensory inputs with high-level decision-making in complex environments. My prior work spans dense correspondence learning, neural implicit 3D reconstruction, and differentiable sensing models developed in challenging domains, which inform my current focus on generalizable perception-planning systems. I also serve as a Graduate Student Instructor for ROB 330 (SLAM & Navigation) and am deeply invested in teaching, mentorship, and outreach.

Home  /  Blog  /  Email  /  CV  /  Scholar  /  LinkedIn  /  Github

PrePrint

Preprints and papers currently under review or in preparation.

MARVO is a physics-aware visual inertial odometry framework for underwater robots that combines a transformer-based feature matcher with a Physics Aware Radiance Adapter to compensate wavelength dependent attenuation, and fuses these corrected visual factors with IMU and barometric depth in a GTSAM factor graph. It further applies a reinforcement learning based pose graph optimizer on SE(2) to refine global trajectories, improving robustness in turbid, visually degraded underwater environments.

Selected Projects

A collection of projects I've worked on, ranging from reinforcement learning to neural scene representations.

Website template based on Jon Barron's academic website. Feel free to use this source code for your own academic portfolio.