Title: "Sense, Shape, and Skill: From Minimal Structures to Practical Robot Capabilities"

Maks Sorokin

Robotics Ph.D. Student

School of Interactive Computing

Georgia Institute of Technology

Date: Friday, August 1st, 2025

Time:  9-11am ET

Location: (Virtual) https://gatech.zoom.us/j/94601071486

Committee 

Dr. Sehoon Ha (Advisor) - School of Interactive Computing, Georgia Institute of Technology

Dr. Danfei Xu - School of Interactive Computing, Georgia Institute of Technology

Dr. Sonia Chernova - School of Interactive Computing, Georgia Institute of Technology

Dr. C. Karen Liu - Department of Computer Science, Stanford University

Dr. Jie Tan - Senior Staff Research Scientist, Google DeepMind

Dr. Simon Le Cleac'H - Research Scientist, RAI Institute

Abstract

Autonomous robots must navigate complex environments, manipulate objects, and adapt to new tasks - yet current approaches often require massive datasets and extensive training for each capability. This thesis demonstrates that practical robot capabilities emerge not from processing every possible signal, but from purposefully structured learning and systematic knowledge reuse.

I present three contributions: (1) semantic navigation methods that strip visual input to essential boundaries, enabling policies trained in abstract simulations to successfully navigate real urban sidewalks; (2) morphology-aware design revealing that robots maintaining visual access to task-relevant features learn 25× more efficiently than those optimized for mechanical metrics alone; and (3) proposed work on athletic manipulation investigating how to achieve fast and steerable policy learning for whole-body control.

By identifying essential representations, designing for information preservation, and structuring for reusability, this thesis offers a path toward more efficient and deployable robot learning systems.