I am a postdoctoral fellow at Columbia working on AI, operations, and robotics. I combine foundation models, control, and optimization to develop autonomous systems, including automated vehicles, robotic manipulators, and AI agents. I received my Ph.D. in EECS from UC Berkeley and previously interned at NVIDIA.
I am best reached at fywu85 (at) gmail (dot) com. You can also find me on Google Scholar and LinkedIn.
Efficient Optimization of Autonomous Agents with Differentiable Language Models
Scheduled · · Session 1:15–2:30 PM PT · Moscone South-312 (Level 3)Control of Multi-Agent Dynamical Systems with Applications to Mixed-Autonomy Transportation
Learning to Accelerate: A Method for Speeding Up Costly Computation
Control of Multi-Agent Dynamical Systems with Applications to Mixed-Autonomy Transportation
Introduction to Connected and Automated Vehicles and Field Experiments
Control of Multi-Agent Dynamical Systems in Understructured and Hyperstructured Environments
Control of Multi-Agent Dynamical Systems with Applications to Mixed-Autonomy Transportation
From Distributed Control to Centralized Optimization: Charting the Next Decades of Autonomous Driving
From Distributed Control to Centralized Optimization: Charting the Next Decades of Autonomous Driving
An MPC Approach to Longitudinal Trajectory Planning
Motion Planning in Understructured Road Environments with Stacked Reservation Grids
Navigating through Chaos: Planning with Stacked Reservation Grid
An Open Robotics Platform for Multiagent Learning and Control
Connections between Car-Following Models and Neural Networks
A Tutorial on Simulation of Urban Mobility
Hybrid Microscopic Traffic Modeling Using a Classic Car Following Model and a Corrective Neural Noise Model
Fuel Consumption in Oscillatory Traffic: Experimental Results
My research spans the areas below. See the full publication list.
I study how robots and AI agents can learn, reason, and act, with a focus on differentiable latent-space models and memory for vision-language-action (VLA) models. See all papers in this area.
Recent Impact: I contributed to Agents’ Last Exam, a benchmark adopted by frontier labs to evaluate their models, including GPT-6 Astra, Opus 5, and Gemini 3.1 Pro.
I combine optimization, control theory, and learning to enable efficient planning, stable control, and coordination among autonomous agents. See all papers in this area.
Recent Impact: I helped KNAPP, a global warehouse automation leader, achieve 80% higher simulated throughput than its recorded production benchmark.
I integrate sensing, hardware, and control to develop and evaluate autonomous systems, with an emphasis on autonomous driving and field experiments. See all papers in this area.
Recent Impact: I contributed to the adaptive cruise control feature in NVIDIA DRIVE AV, which is used in Mercedes-Benz production lines.