CV

Contact Information

Name Alenna Spiro
Professional Title PhD Student, Computer Science
Email spiro.a@northeastern.edu

Professional Summary

Computer Science PhD student at Northeastern University, working in the Northeastern Autonomy and Intelligence Laboratory with Prof. Michael Everett. My research combines reinforcement learning, control theory, and decision-making under uncertainty to make autonomous systems more reliable when they fail, drawing on models of human learning and development for more efficient robot learning.

Experience

  • 2024 - present

    Boston, MA

    Graduate Research Assistant
    Northeastern University
    Northeastern Autonomy and Intelligence Laboratory, advised by Prof. Michael Everett. Research on reliable decision-making and learning-based control for autonomous systems (see Selected Research).
  • 2022 - 2024

    Amherst, MA

    Graduate Research Assistant
    University of Massachusetts Amherst
    Autonomous Learning Laboratory, advised by Prof. Bruno Castro da Silva (see Selected Research).
  • 2023 - 2023

    Amherst, MA

    Teaching Assistant
    University of Massachusetts Amherst
    • Teaching assistant for two graduate courses under Prof. Bruno Castro da Silva: COMPSCI 589 (Machine Learning, Spring 2023) and COMPSCI 687 (Reinforcement Learning, Fall 2023); graded hundreds of assignments and exams.
  • 2019 - 2022

    Amherst, MA

    Undergraduate Research Assistant
    University of Massachusetts Amherst
    Laboratory for Perceptual Robotics (advisor: Prof. Roderic Grupen).
    • Belief-space object recognition and mobile manipulation in the Roger simulator; second author on the resulting publication (ICDL 2022).

Education

  • 2024 - 2029

    Boston, MA

    PhD
    Northeastern University
    Computer Science
    • Khoury College of Computer Sciences.
    • Northeastern Autonomy and Intelligence Laboratory (advisor: Prof. Michael Everett).
  • 2022 - 2024

    Amherst, MA

    MS
    University of Massachusetts Amherst
    Computer Science
  • 2019 - 2022

    Amherst, MA

    BS
    University of Massachusetts Amherst
    Computer Science

Selected Research

  • ENCORE: Environment-aware Cost-optimal Fault Recovery

    First author (Spiro & Everett) · in preparation

    • Built a recovery advisor for autonomous ground vehicles that decides whether a detected fault warrants physical repair or continued degraded operation by minimizing total mission cost, rather than the fixed safety thresholds used by prior self-assessment methods.
    • Designed a learned, per-component repair-effectiveness model and a risk-sensitive (CVaR) action selector that weighs intervention cost against predicted post-repair performance, generalizing across fault types and compounding faults without fault-specific retraining.
    • Evaluated in a custom differential-drive simulation across four hardware fault types and varied environments against do-nothing and full-repair baselines.
  • A Hybrid Framework for Efficient Koopman Operator Learning

    Co-first author (equal contribution) · CDC 2025 · delivered the oral presentation

    • Combined semidefinite programming with deep learning to model nonlinear dynamical systems as linear Koopman operators.
    • Used a small-scale SDP to derive the observable-space dimension, system order, and an approximate Koopman operator, then used that structure to initialize and train an autoencoder, removing hyperparameters that learning-only methods must otherwise guess and avoiding the quadratic scaling of SDP-only methods.
    • Demonstrated lower prediction error and faster convergence than a learning-only baseline across four dynamical systems, including the chaotic Lorenz attractor.
  • SPARTA: Smooth Point-cloud Approach-angle Reasoning for Terrain Assessment

    Contributing author (4th of 7) · CoRL 2025

    • Estimates angle-of-approach-dependent terrain traversability from point clouds, predicting a smooth Fourier-basis risk function over approach angle that can be queried cheaply during planning instead of re-running inference per angle.
    • Improved boulder-field crossing success to 91% versus 73% for an elevation-based baseline in high-fidelity simulation, with hardware validation.
  • SACRED: Structure and Affordance-based Categorization of Related Environmental Descriptors

    Master’s research · UMass Autonomous Learning Lab (advisor: Prof. Bruno Castro da Silva)

    • Trained a variational autoencoder to reconstruct environment images, then built per-action, class-conditional histograms over each latent dimension from successful versus failed executions (success defined as an effective change in the environment) — a generative model of how likely each action is to be effective in a given state.
    • Scored candidate actions at decision time by encoding the current view into the VAE latent space and evaluating each action’s success log-likelihood under this affordance model, surfacing the actions a state most affords.
    • Used those affordance scores to guide exploration in a Q-learning agent navigating a room-based, image-observation world (doors, walls, cabinets, with open/close/move/turn actions), biasing exploration toward effective actions and substantially accelerating learning over undirected exploration.

Publications

Honors and Awards

  • 2024
    Khoury Distinguished Fellowship
    Northeastern University
  • 2022–2024
    Bay State Scholarship
    University of Massachusetts Amherst
  • 2019–2022
    Chancellor's Award
    University of Massachusetts Amherst
  • 2019
    NASA Massachusetts Space Grant Consortium Research Grant

Skills

Programming Languages: Python, C/C++, Java, JavaScript, C#, MATLAB
ML / RL: PyTorch, TensorFlow, Gymnasium, Weights & Biases, NumPy, SciPy, pandas, scikit-learn, OpenCV, CVXPY
Robotics: ROS/ROS2, Gazebo, MuJoCo, custom simulator development
Tools: Git, Docker, LaTeX, Linux (Fedora, Ubuntu)

Languages

English : Native
Spanish : A2
Arabic : A1 (Modern Standard, Levantine, Egyptian; ongoing)