Photo

Alexander Alexiev

I am an undegraduate student studying Engineering Science at the University of Toronto. I am pursuing a Major in Robotics with a Minor in Artificial Intelligence. Throughout my degree, I worked at Electronic Arts as a Machine Learning Intern, I worked at MIT as a Biomedical Robotics Researcher, and I am currently an undergraduate researcher at the Robot Vision and Learning group at the University of Toronto, supervised by prof. Florian Shkurti.

My goal is to create robots that interact with the world as effortlessly as humans do. I aim to achieve this by conducting research at the intersection of computer vision, mechanical design, electrical engineering, and artificial intelligence.

   /   alex.alexiev [at] mail (dot) utoronto (dot) ca



Publications and Patents

Hen-Wei Huang, Peter R. Chai, ..., Alexander Alexiev, and Giovanni Traverso
An Implantable System for Opiod Safety
Device 100517 [Paper] [Press]
Alexander Alexiev, Jack Chen, Giovanni Traverso
Ingestible Light-Reflectance Sensing Capsule for Automated Ischemia Detection
Under Review at Science Robotics
Alexander Alexiev, Jack Chen, Giovanni Traverso
Electrically-Triggered Microactuator Powered by Aluminum-Nickel Nanofilm
U.S. Provisional Patent


Projects

Real-time Non-Rigid Body Tracking and Manipulation
A novel approach to tracking and manipulating deformable objects, like ropes. Rather than Gaussian splatting, a computationally expensive method, I use point clouds, which are extremely fast and enable real-time state estimation of non-rigid bodies. I apply the non-rigid point set registration technique to keep the object's points consistent across all frames. I successfully trained a graph neural network to model the rope’s dynamics.
Task and Motion Planning with Object Relevance
A novel machine learning architecture, encoding visual data with a convolutional neural network and scene graph information with a graph neural network. This system predicts object relevance scores to help direct the search process in task and motion planners (specifically a PDDLStream variant).
Enhancing Time Series Prediction Through Short-Term Covariate Forecasting
A framework that is able to first forecast covariates into the near future and then use those predictions along with past known covariates to estimate the current target variable.
Electrically-Triggered Jetting of mRNA Vaccine Into Internal Tissue
Designed and created a novel ingestible robot capable of delivering mRNA vaccines directly into the stomach tissue, thus alleviating the need for an injection.
Ultra-Low Power Microheater
Developed a novel ultra low power method to ignite a reactive metallic nanofilm on demand using a custom micro-heater array.
Mesoscale Single-Cut Bistable Mechanism
Created the first mesoscale (1-500 um) bistable mechanism that can be cut in a single laser cutting operation out of Nickle chromium alloy. Can be used in a variety of applications, such as drug delivery, micro-robotics, and MEMS.
RoboSoccer
Designed a multi-agent control algorithm using decision trees for a team of 4 humanoid robots playing soccer for the RoboCup 2022 in Thailand.
University of Toronto Formula Racing Team
Created the wiring harness for the car and designed a cell-level fusing system for a custom 600V Li-ion battery pack made with 18650 cells.
Ball Balancing Robot
Created a robot that balances on a ball by using the digital signal processor on the IMU for fast closed loop control with a dynamics model and a custom stepper motor controll library.
Actuating Plotter
Built and programmed a plotter for under $15 capable of taking any PNG image and converting it to a series of line segments that can be drawn by two stepper motors.
FIRST Robotics Competition Mentor
I started my high school's FRC team in grade 11. I've since continued as a mentor teaching modern robotics methods and tools to high school students for the FIRST Robotics Competition. Currently using Nvidia Isaac Sim with ROS2 to develop and test a highly optimized C++ library using multiple cameras and AprilTag targets for fast and precise localization with an extended kalman filter and factor graphs.
Drones
In the 8th grade I designed, 3D printed, and programmed on an Arduino multiple drones with custom gimbals and GPS navigation. Fly FPV drones recreationally.


Experience

Robot Vision and Learning Lab @ University of Toronto


  • Currently working on novel methods for robots to manipulate ropes in 3D.
  • Created a computer vision pipeline to incorporate learning into task and motion planning (TAMP) problems by learning representations for 3D scenes with a series of CNNs and GNNs using perception data and automatically generated scene graphs from object poses.
  • Used Slurm and Docker to parallelize multiple data collection instances from a DRAKE simulation of a Franka Panda environment and multiple model training sessions in Tensorflow to gather training statistics and compare different models efficiently.

Lab for Translational Engineering @ Massachusetts Institute of Technology and BWH


  • Led the development of a pill sized ingestible robot to autonomously deliver mRNA vaccines inside the stomach through a non-invasive manner. (Manuscript under preperation)
  • Created an ingestible robot to create a map of the small intestine in 10 wavelengths to non-invasively and autonomously detect internal bleeding and bruising. (Manuscript under preparation)
  • Developed a low power algorithm in C++ to fuse sensor data on an implant and detect opioid overdoses. An Implantable System for Opioid Safety, Device 100517

Machine Learning Team @ Electronic Arts


  • Created a tool for the Frostbite game engine in C++ to gather 3D bounding boxes for key animation objects and stream the data to a remote server for machine learning models.
  • Optimized the FIFA game testing process by designing an Autoencoder CNN with Tensorflow to detect visual animation glitches and therefore allowing human game testers to focus on more complex feedback.