Learning from real-world traffic toward Reliable Autonomy

I’m actively looking for industrial and academic opportunities in
Autonomous Driving and Embodied AI.

Hello, I am Weijiang Xiong (熊伟江), a Ph.D. candidate in Robotic Systems at EPFL (EDRS), advised by Prof. Nikolas Geroliminis and co-advised by Prof. Alexandre Alahi.

My research focuses on learning from real-world traffic, with the goal of developing autonomous systems that accounts for uncertainty and adapts to changing environments. I develop probabilistic models of traffic dynamics and study how aerial traffic observations can support autonomous driving. My recent work turns aerial urban traffic monitoring into driving scenarios and supervision for adapting driving models to new cities. I am also exploring reinforcement learning for active urban sensing.

My earlier work in visual perception, Bayesian machine learning, and domain adaptation shaped my interest in the reliability of learned systems. During the journey, I had the pleasure to work with Dr. Lorenzo Bertoni, Dr. Taylor Mordan and Dr. Markus Heinonen. My longer vision is to learn autonomous driving and its environment, i.e., the transportation system, by connecting traffic observation, behavior learning, and adaptive decision-making.

Research interests

Probabilistic Modeling
Cross-City Adaptation
Active Sensing

Selected publications

My work spans probabilistic traffic forecasting, learning from aerial observations, and understanding road users’ behavior.

arXiv 2026

SkyDrive workflow turning aerial traffic observations into supervision for autonomous driving

Weijiang Xiong, Lan Feng, Alexandre Alahi, Nikolas Geroliminis

Preprint

SkyDrive: Learning to Drive in a New City from Aerial Traffic Monitoring

Uses aerial traffic observations to help autonomous driving models adapt to new cities.

arXiv preprint, 2026

TR-C 2026

GMM layer adapting a traffic forecasting network to predict probability distributions

Weijiang Xiong, Robert Fonod, Nikolas Geroliminis

Preprint Code Data

Unveiling Stochasticity: Universal Multi-modal Probabilistic Modeling for Traffic Forecasting

Turns traffic forecasting networks into probabilistic predictors with a Gaussian Mixture Model output layer, capturing multiple possible future traffic states and their uncertainty.

Transportation Research Part C: Emerging Technologies, 2026

IEEE T-ITS 2025

HiMSNet architecture combining drone observations and loop detector measurements

Weijiang Xiong, Robert Fonod, Alexandre Alahi, Nikolas Geroliminis

Paper Code Data

Multi-Source Urban Traffic Flow Forecasting With Drone and Loop Detector Data

Combines loop detector measurements and drone observations through a hierarchical neural network for urban traffic forecasting, accompanied by a multi-source dataset built from vehicle trajectories.

IEEE Transactions on Intelligent Transportation Systems, 2025

Signal Processing 2020