CONTACT
Book an Appointment
© 2025 Roman Aguilera
Roman is the Artificial Intelligence and Machine Learning Engineering Lead at PivotalVC, where he oversees the development of AI- and ML-powered web applications, leveraging cutting-edge models to build enterprise-grade solutions.
Roman was an AI Robotics Researcher and Egnineer at the UCSB Dynamic Robotics Laboratory for 5 years, where he investigated fundamental AI algorithm performance on robot control tasks. He created custom simulation environments of highly redundant N-link robot arms using Open AI’s Gym API in order to evaluate control algorithm performance as it was affected by (i) the number parts on a robot, (ii) the complexity of a robot’s environment, and (iii) the complexity of a robot’s task. He ultimately discovered evidence to suggest that the Proximal Policy Optimization algorithm learns motions rather than making sense of end goal points. His work and contributions were awarded the Google-CAHSI Dissertation Fellowship in 2023.
Roman was an Computer Vision Engineer at a confidential robotics startup where he contributed to an existing vision system and automated the collection of plant data. He modified the electrical wiring of their vision system in order to mount it onto their field tractor robot. He also automated the implementation of image recognition algorithms to detect center points of strawberry plants and record the GPS position of plant center points during plant treatment.
Roman was an Embedded Systems Researcher and Engineer at the UCSD Neural Interaction Laboratory for 2 years. He worked with the Programmable-System-on-Chip 4 BLE, leveraging the onboard modules such as the analog-to-digital converter, operational amplifier, and Bluetooth-Low-Energy transmitter to create a prototype device that transmits epidermal sensor information to a smartphone application and computer. He programmed the microcontroller in C, wired the electrical components, and also redesigned the electrical circuits. His work contributed to a larger project that was published in Advanced Materials 2017.
This project addresses the challenge of designing controllers for high-dimensional robotic systems while ensuring computational tractability. Discretizing the robot’s state space (e.g., a 3-link robotic arm) leads to an exponential increase in possible states, making brute-force methods impractical due to memory and time constraints. We overcome this by using first principles to determine limits on the mesh of reachable states. We also implement Barycentric Interpolation in order to allow information flow when implementing Value Iteration on dynamical systems models.
This project aims to build end-to-end, hard real-time, control systems (namely agile humanoids) by integrating open-source software tools. The end result was a custom real-time Operating System for robotic control using Xenomai & Ubuntu.
In this project, I used Tableau to determine high- and low- performing advertisement groups based on spatial data. The metrics analyzed were CVR, CTR, Viewability Rate, Impressions, Conversions, CPC, CPM, CPA, and CPvM.
Ph.D. Computer Science (2024) — UC Santa Barbara (All But Dissertation)
M.S. Computer Science (2024) — UC Santa Barbara
B.S. Electrical Engineering (2017) — UC San Diego
Robotic Learning, Manipulation, Locomotion, Humanoids, Real-Time Control, Embedded Systems, AI, Operator Theory