Years of experience
Roman Aguilera
Santa Maria,CA
Summary
Innovative robotics and AI professional with extensive experience in developing advanced reinforcement learning algorithms at UCSB Robotics Lab. Demonstrated success in enhancing robot locomotion and manipulation through collaborative projects, leveraging strong skills in Python and machine learning. Committed to transforming theoretical concepts into practical applications that drive technological advancement.
Overview
5
5
Work History
Field Data Science Intern
TRIC Robotics
09.2024 - 11.2024
- Trained a sophisticated neural network architecture for precise detection of geometric center points in video game frame analysis of wooden blocks configurations
- Set up and calibrated the camera system in the office and field, ensuring proper image quality and connectivity with other robot hardware
- Collected image data according to project guidelines, maintaining organization and documentation
- Made modifications to data collection software to improve ease of use
AI Robotics Researcher
UCSB Dynamic Robotics Lab
05.2018 - 06.2024
- Developed, trained, and deployed reinforcement learning algorithms for locomotion and manipulation tasks
- Built simulation infrastructure to support the training of locomotion and manipulation policies for a general purpose humanoid robot at a large scale
- Collaborated with the controls team to integrate policies into the existing control stack
- Defined, tested, and evaluated performance metrics for learned policies
- Wrote production quality code in PyTorch for 5 years
- Worked with online and offline reinforcement learning algorithms such as PPO, SAC, TRPO, and TQC+HER
- Tuned hyperparameters and cost functions for the RL algorithms
- Implemented common RL techniques such as domain randomization, curriculum learning, and reward shaping
- Trained locomotion policies for quadrupedal robots and bipedal robots
- Created evaluation tools to evaluate Reinforcement Learning Policies
- Investigated fundamental performance of control algorithms, as robot parts and simulation environment were changed
- Investigated performance of Rapidly-Exploring Random Trees and Model-Free/Model-Based Reinforcement Learning
- Discovered evidence to suggest that the PPO algorithm is learning motions, rather than making sense of goal points
- Video demonstration of results here https://drive.google.com/file/d/1UX_fEvDkoU-bSyNF386ugiMh_kiHxJVL/view?usp=drive_link
- Research proposal here https://docs.google.com/document/d/16SNo08ZLLnMyZ6FgPo8fR3_Y8aGxxpDIAwFg4jhn-
- Successfully trained a 32-link arm control policy such that the end effector would touch a goal point
- (Python, OpenAI Gym)
- Developed a Python script that automatically generates a URDF/XML model of a robot’s physical properties, for an arbitrary number of links desired on a robot
- URDF model was used in a simulation environment
- (Python, xml.etree.ElementTree Python Package)
- Created over 8 custom reinforcement learning simulation environments for a multi-link robot arm
- (Python, OpenAI Gym, OpenAI Baselines, Stable-Baselines 3, PyBullet Physics Simulator, Mujoco Physics Simulator)
- Applied PPO algorithm to perform policy search for multilink arm control
- (Python, OpenAI Baselines, Stable Baselines)
- Applied RRT algorithm to perform trajectory search for multi-link robot arm control, such that the end effector reaches a goal point
- The code works for an arbitrary number of links
- (MATLAB)
- Implemented Value Iteration algorithm with Barycentric Interpolation on both Grid World and Double Integrator control problems
- (MATLAB)
- Performed literature reviews on Reinforcement Learning, Koopman Operator Theory, and Trajectory Optimization
Teaching Assistant
UCSB Computer Science/ UCSB ECE / UCSB Physics
09.2020 - 03.2024
- Provided comprehensive instructional leadership for diverse technical courses spanning robotics, computer science, and physics disciplines for 8 courses
- Created solutions for homework and lab assignments, hosted sections and office hours, and graded assignments
Research Mentor
UCSB CSEP / UCSB EUREKA Scholars Program
06.2020 - 08.2020
- Supervised undergraduate mentee in their summer research project
- Taught basic concepts in Object-Oriented Programming, Python, Reinforcement Learning, and PyBullet
- Met with undergraduate mentee at least 2 times per week to ensure adequate progress
Education
M.S. - Computer Science, Research in Robotics and Machine Learning
University of California, Santa Barbara
Santa Barbara06.2024
B.S. - Electrical Engineering, Depth in Machine Learning and Controls
University of California, San Diego
San Diego12.2017
Skills
- C
- C
- Python
- MATLAB
- Racket
- Rosette
- R
- PSPICE
- MIPS R2000 Assembly
- Xenomai
- Ubuntu
- Linux
- Vim
- Roboflow
- OpenAI Gym
- OpenAI Baselines
- Stable-Baselines 3
- RLlib
- PyBullet
- Mujoco
- PyTorch
- TensorFlow
- CVXPY
- Model Free Reinforcement Learning
- Rapidly Exploring Random Trees
- Policy Optimization Techniques
- Support vector machines
- Feature engineering
- Predictive modeling
- Recurrent neural networks
- Linear regression
- Logistic regression
- Scikit-learn
- Network analysis
- Bayesian statistics
- Data modeling design
- Optimization algorithms
- Dimensionality reduction
- Time series analysis
- Deep learning
- Simulation modeling
- R programming
- Principal component analysis
- K-means clustering
- Model evaluation
- Neural networks
- Rapid application development (RAD)
- Python programming
- Regularization techniques
- Machine learning
- Image processing
- Graph theory
- Convolutional neural networks
Publications
Yun-Soung Kim, Jesse Lu, Benjamin Shih, Armen A. Gharibans, Zhanan Zou, Kristen Matsuno, Roman Aguilera, Yoonjae Han, Ann Meek, Jianliang Xiao, Michael Thomas Tolley, Todd P. Coleman, Scalable manufacturing of solderable and stretchable physiologic sensing systems, Advanced materials, 29, 39, 2017
Googlescholar
https://scholar.google.com/citations?user=DS1I_BUAAAAJ&hl=en
Languages
Spanish
Professional
Accomplishments
Google-CAHSI Dissertation Fellowship ($25,000)
NSF-IGERT Network Science Fellowship ($30,000)
NSF-LSAMP Bridge to Doctorate Fellowship ($64,000)
Timeline
Field Data Science Intern
TRIC Robotics
09.2024 - 11.2024
Teaching Assistant
UCSB Computer Science/ UCSB ECE / UCSB Physics
09.2020 - 03.2024
Research Mentor
UCSB CSEP / UCSB EUREKA Scholars Program
06.2020 - 08.2020
AI Robotics Researcher
UCSB Dynamic Robotics Lab
05.2018 - 06.2024
M.S. - Computer Science, Research in Robotics and Machine Learning
University of California, Santa Barbara
B.S. - Electrical Engineering, Depth in Machine Learning and Controls
University of California, San Diego
Similar Profiles
Dave WatsonDave Watson
Inside Technical Sales at Abbey Mec-TricInside Technical Sales at Abbey Mec-Tric
Dave CraftsDave Crafts
Engineering R&D at Tric ToolsEngineering R&D at Tric Tools
Sneha Ganesh NayakSneha Ganesh Nayak
Robotics Software Engineer at Path Robotics, Inc. Path Robotics, IncRobotics Software Engineer at Path Robotics, Inc. Path Robotics, Inc