Robotic Inspection Apparatus for Lockheed Martin
For my undergrad capstone, I co-led an effort to produce an inspection rig for the F-35 assembly line. The purpose of the fixture was to reduce setup time for inspection and water-jet stress testing of composite fuselage panels.
​
Our design, which leveraged a robotic arm and linear actuators to dynamically resize itself to grip panels of diverse geometries, reduced total inspection time by up to 85%.
​
Our team won the Sally Blum Memorial Prize, given annually for the most outstanding capstone project in Mechanical Engineering at Southern Methodist University.
​
My contributions included ownership of the actuator subassemblies, from design through fabrication and testing. Additionally, I spearheaded a holistic redesign focused on DFM, simpler parts, and material reduction that cut overall cost by >50%. For example, using FEA, I found that the diameter of the rails supporting our screw drives could be halved while maintaining an adequate factor of safety. That discovery alone saved us 5% of our budget.
One of my favorite design contributions was the two-beam approach I came up with for keeping the top plane of the robotic arm horizontal. I got the idea from boat stairs that use a parallelogram geometry to maintain their orientation regardless of angle, as shown in the image.
Through this project, I built experience in every facet of mechatronic design, including using mechanical, electrical, and computational methods simultaneously to debug the system. I gained a bias for action and an instinct to put fires out as they arose.
Forward Sensor Module for Torc Robotics
This image shows the sensor module I designed for the forward-facing LIDARs, microphones, and camera that help a Torc truck drive itself. I owned the design of all of the sensor mounting hardware in that photo.
​
I was especially content with the plastic shell on the underside which houses the microphones and camera. I designed it to be 3D-printed or injection molded, and it included some cool geometries for flush mounting, as well as to drain rainwater.
​
Clearly, Torc Robotics was happy with my work, as they included my design in their promotional material announcing an extension of their partnership with Aeva.
​
I designed that assembly while working in Blacksburg, Virginia. That photo was taken in Mountain View, California. It's pretty cool to think that the work I did as an intern has helped Torc trucks drive themselves across the country.
Inverse Kinematics Using Neural Networks
For the final project in my machine learning class, my group employed multilayer perceptron neural networks to classify and predict the inverse kinematic solution space for redundant robotic manipulators. Our project was chosen by the faculty as the most outstanding of the semester, and is showcased on the course website.
​
Inverse kinematics is the process of solving for a robotic arm's joint angles from a given end-effector position. For high degree-of-freedom systems, geometric solutions are computationally expensive at runtime. We implemented a supervised learning model on a large synthetic data set to classify the quality of the potential IK solutions in a given area, and another model to allow the controller to infer the IK solutions at runtime instead of calculating them.
​
I personally implemented the classifier using scikit-learn and an innovative KDTree approach using scipy for data preprocessing. I also built the first iteration of the IK solver model. Our build was both more accurate and significantly faster at runtime than typical numerical methods.​
​
The GitHub repository can be found here, and you can read the paper here.