Taekyung (TK) Lee

EE (Intelligent Systems)+ Robotics + Control & Dynamical Systems (CDS)

Passionate about multi-agent systems, learning-based control, and reinforcement learning

Contact Me LinkedIn GitHub
Taekyung Lee Profile Picture

About Me

I am an Electrical Engineering student at the California Institute of Technology (Caltech), pursuing the Intelligent Systems track with a minor in Robotics, Control & Dynamical Systems. My research interests lie at the intersection of multi-agent systems, learning based control, and reinforcement learning.

Most recently, I was a Frederick W. Drury, Jr. SURF Fellow at Caltech's Autonomous Robotics and Control Lab (ARCL), developing multi-agent fault detection and isolation frameworks using graph neural networks. Also, I was a Robotics Institute Summer Scholar (RISS) at Carnegie Mellon University, working on spatiotemporal risk-aware planning for autonomous navigation in hazardous environments.

My research experience spans learning-based adaptive control, model predictive path integral control, and autonomous exploration with SLAM. I am also a proud recipient of the Korean Presidential Science Scholarship.

Pasadena, CA
Expected Graduation: June 2027
GPA: 4.2/4.3

Research

Jun 2025 - Present

Frederick W. Drury, Jr. SURF Fellow - Multi-Agent Fault Detection & Isolation

Autonomous Robotics and Control Lab, California Institute of Technology

Advised by Professor Soon-Jo Chung

Pasadena, CA

  • Developed two-stage local FDI framework: (1) dual VAE/CVAE architecture learning latent representations of subsystem telemetry with causal modeling to distinguish sensor faults (VAE) from actuator faults (CVAE conditioned on parent sensor states), (2) MLP classifier with Stiefel manifold orthogonality constraints on concatenated latent features
  • Implemented global consensus module using GraphSAGE graph neural network for distributed belief aggregation across agent neighborhoods, improving fault detection accuracy through multi-hop message passing up to 50-agent swarms
  • Extending framework on realistic multi-spacecraft scenarios using Basilisk astrodynamics simulation, developing state estimation pipeline and recovery strategies for communication link failures, preparing manuscript for submission
SURF Presentation View Demo View Presentation
Jun 2025 - Present

Robotics Institute Summer Scholar - Spatiotemporal Risk-Aware Proactive Planning

Advanced Agent & Robotics Technology Lab, Carnegie Mellon University

Advised by Professor Katia Sycara

Pittsburgh, PA

  • Developed risk-aware planning framework for autonomous navigation in spatiotemporally evolving hazardous environments using CNN-ConvLSTM architecture for model-free multi-horizon environmental forecasting
  • Implemented adaptive MPPI controller with CVaR-incorporated risk optimization balancing direct damage avoidance and worst-case exposure, achieving 40% cumulative risk reduction vs. baseline safe-set methods and preventing local minima trapping in wildfire simulation
  • Establishing baselines against CBF, HJ-reachability, and Safe RL methods; developing time-series diffusion model for enhanced online prediction accuracy toward Robotics and Automation Letters (RA-L) submission (Nov 2025)
RISS Presentation RISS Research Poster View Video
Jan 2025 - Aug 2025

Student Researcher - Learning-Based Adaptive Control for Quadrotor Table Tennis

Autonomous Robotics and Control Lab, California Institute of Technology

Advised by Professor Soon-Jo Chung

Pasadena, CA

  • Implemented PPO-based controller for quadrotor ball interception in table tennis environment using Omnidrones
  • Designed hierarchical control architecture with high-level strategic planner selecting optimal skills from imitation-learned policy sets based on real-time opponent movement analysis and ball trajectory prediction
  • Developed physics-informed neural network modeling time-varying quadrotor-gimbal-paddle dynamics with coupling inertia parameterization enabling stable control under dynamic center of mass conditions