Nikita Sarawgi

I'm a robotics engineer at GrayMatter Robotics, working on closed-loop active perception and decision-making for robots on industrial shopfloors.

I completed my M.S. in Computer Science at USC Viterbi in December 2025, advised by Satyandra K. Gupta and Daniel Seita.

My research is motivated by a simple problem: robots deployed in the real world routinely encounter conditions outside their training distribution. My goal is to build systems that adapt to this uncertainty rather than fail silently, by composing learned behaviors and representations to generalize across novel environments and task variations.

My work spans perception, manipulation, and sequential decision-making in real-world settings. I am interested in how robots can use perception to actively reduce uncertainty before acting, and how decision-making systems can select and sequence behaviors to complete tasks even as conditions shift at test time.

Email  /  Resume  /  Scholar  /  LinkedIn  /  Github

Updates

Publications

Research at the intersection of robot learning, active perception, and reasoning under uncertainty. Selected work below.

An RL policy for online 3D bin packing that chooses both the next item to pack and the face to grasp it on. A preference parameter set at inference time trades off packing density against operation time without retraining.

An autonomous insertion system for in-space assembly. The robot first estimates a coarse pose to position a wrist-mounted camera near the target, then uses a learned model to estimate fine misalignment, and executes the insertion under impedance control.

Website adapted from here.