Keuntaek Lee

Papers

Total Citations

H-Index

About

Keuntaek Lee is a researcher at the forefront of integrating perception, control, and reinforcement learning for autonomous systems. His work centers on developing robust decision-making frameworks for robots operating under uncertainty, with a particular emphasis on vision-based navigation and risk-aware policy optimization. Lee's most impactful contribution is his pioneering work on "Aggressive Perception-Aware Navigation Using Deep Optical Flow Dynamics and PixelMPC" (2020, 37 citations), where he introduced a novel coupling of model predictive control with deep optical flow to enable high-speed, agile flight in cluttered environments. This work directly addresses the challenge of fusing visual data with robot dynamics for real-time control. He has also made significant strides in reinforcement learning, proposing a "Sample-based Distributional Policy Gradient" (2020, 7 citations) that captures the intrinsic randomness of long-term returns, moving beyond traditional expected-value approaches. Furthermore, his framework for "Adaptive CVaR Optimization for Dynamical Systems" (2020, 3 citations) provides a principled method for handling uncertainty from initial conditions and stochastic dynamics, directly tackling the safety-critical aspects of autonomous navigation. Through these contributions, Lee is shaping a new generation of algorithms that are both perceptually aware and statistically robust.