Olumide Ogunmakinwa

Olumide Ogunmakinwa

Biography

Olumide T. Ogunmakinwa is a rising junior and Karsh STEM Scholar at Howard
University, pursuing a bachelor’s in Computer Engineering. His research interests focus on
computer architecture, VLSI design, and the hardware foundations that enable and constrain
intelligent computing systems. As a Karsh STEM Scholar, Olumide is passionate about
expanding access in STEM fields. He serves as a role model for students aspiring to pursue
research and graduate study. He plans to pursue a Ph.D. in Computer Engineering to advance
computing technologies from the ground up, believing that meaningful innovation requires
understanding systems at their most fundamental level. Outside of his academic pursuits and
community involvement, Olumide enjoys gaming and exploring superhero narratives, which
continue to inspire his creative approach to problem-solving in the technology field.

From Human Touch to Robot Hands: An Agentic Pipeline for Processing Multimodal
Tactile Manipulation Data
Olumide Ogunmakinwa1, Y. Ray Song2, and Paul P. Liang2
1Department of Electrical Engineering and Computer Science, Howard University
2Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology

Robots still handle objects far less capably than people, largely because most robot learning uses
vision alone and never sees the forces a hand applies. Wearable tactile gloves can now record those
forces during ordinary activity, and OpenTouch pairs them with hand pose and first-person video
across hundreds of everyday manipulation sessions. Raw recordings are not training data. Each session
must be checked for quality, synchronized, and segmented into individual moments of contact,
work that does not scale by hand. We present an automated pipeline of specialized software agents
that converts raw multimodal recordings into structured, labeled contact events without manual review.
The agents audit recording quality and timing, detect each contact from the glove’s 256 pressure
points, mark its onset, peak, and release, and pair it with hand shape and wrist motion. Across
139 sessions the pipeline has produced over 5,000 contact events, and exposed a persistent sensor
bias in one glove that was suppressing contact detection, now corrected automatically. Ongoing work
extends the pipeline with automated visual annotation of contacts and a scoring method for identifying
which tactile scenarios are most informative for robot training, a step toward robots that learn
manipulation from human touch.