GE Vernova Advanced Research: Robotics Research Intern

May 2025 - August 2025

Overview

At GE Vernova, I contributed to two key projects:

  • SPEEDWORM—an ARPA-E funded initiative—aimed to develop a compact, truck-portable system for minimally invasive undergrounding of electrical infrastructure, unlike traditional horizontal directional drilling. 
  • Wind Turbine Blade Post-Infusion Inspection System designed to ease worker strain, improve defect detection, and generate training data for AI-assisted inspections. 

Due to proprietary constraints, I can only share publicly available details.


SPEEDWORM: Underground Tunneling Robot

While working on SPEEDWORM, I was responsible for developing a prototype steering mechanism using SolidWorks. I reviewed academic literature and prior GE Vernova research, and after team discussions, we chose a “bent cone” design—based on its proven effectiveness in steering and its use in horizontal directional drilling. The design needed to fit within the existing robot, so I avoided bulky assemblies and electronics. I created a system that converts the forward force from the piston into torque, allowing the robot to steer without added components.

After three design iterations, I built a test stand with a piston setup identical to SPEEDWORM’s. By adjusting piston extension and contraction, the robot’s tip could be turned clockwise or counterclockwise.  I also programmed a spare PLC and selected hardware to stress test the mechanism, measuring its range of motion and failure limits.

Near the end of the internship, I presented the working prototype to ARPA-E representatives, explaining its function and design. 

Wind Turbine Blade Post Infusion Inspection System

My primary responsibility on the team was sensor integration. I began by developing an interface between the robot’s main controller and a drone camera API to enable simultaneous image capture for environmental analysis. This required capturing images along with metadata—distance readings from two sensors, camera ID, and timestamp—within 0.2 seconds, the time it takes the system to move one meter. To reduce image setup time, I disabled the camera’s default auto-refocus except at the start of movement. Although this improved performance, file transfer remained a bottleneck. After reviewing the hardware, I replaced USB 2.0 cables with USB 3.0 to increase transfer speed. I then implemented multithreading to allow all cameras to capture images simultaneously. This reintroduced transfer delays, which I resolved by deferring image transfers until the end of the inspection. Metadata was logged before each image set and later merged with the images prior to saving to the SSD.

My second task involved mechanical design. The original actuator homing method relied on torque sensors, which proved unreliable. I was assigned to develop custom limit switch mounts to detect when the robot was homed. Due to constraints on modifying the system, I designed sensor attachments that required minimal changes. One actuator had a wide range of motion, making a standard limit switch impractical. To address this, I created a flexible lever arm that would trigger the switch when homed. Another actuator lacked space for a switch beneath it, so I designed a small extension arm that reached an open area where the switch could be mounted.

The system was tested at a GE Vernova wind turbine blade factory. Compared to traditional post-infusion inspection methods, workers identified more defects and reported reduced physical strain. Based on these results, the system is now being refined for deployment at additional GE Vernova facilities. 

Company Website

  • GE Vernova