Meng Wang

I'm a Machine Learning Engineer at Waymo, where since January 2025 I've worked on perception models for autonomous driving, improving how the system detects and interprets the world around the vehicle, and how that carries through to end-to-end driving performance.

Before Waymo, I worked on computer vision and machine learning for driving data at scale: detecting edge cases in large-scale naturalistic driving video as a Machine Learning Engineer at MIT AgeLab, and applying ML/CV and LLMs to multi-modal driving data as a Research Intern at Honda Research Institute in 2024.

I received my PhD in Industrial Engineering and Operations Research (IEOR) at the University of Massachusetts Amherst, advised by Dr. Shannon Roberts, where my research applied machine learning and statistical modeling to crash prediction and driving data.

Email  /  Scholar  /  LinkedIn

Research & Publications

My work focuses on perception and machine learning for autonomous driving, large-scale visual data, vision-language models, and driving scene understanding. Representative papers are highlighted.

Machine Learning & Perception

A two-stage framework that integrates semantic, contextual, and kinematic roadway-complexity features for crash prediction, reaching 90.15% accuracy with the complexity-infused features.

A unified deep-learning model that performs keypoint detection and spatiotemporal tracking in a single joint-learning framework, operating on in-the-wild video.

A deep-learning method for clustering common cases and detecting edge cases from the visual characteristics of the external driving scene, plus a large-scale real-world video dataset of driving scenes built with it for evaluating perception systems.

Driver Behavior & Vehicle Automation

A simulator study implementing staged and simultaneous HMIs and analyzing the drivers' driving and glance behavior.

A literature review of the HMI recommendations and vehicle automations.

Evaluating an updated training program (based on RAPT and ACCEL) by introducing a more diverse set of training scenarios and leveraging deployment platforms that were better suited to modern devices.

A simulator study assessing the effectiveness of training and warning systems on drivers' response behavior to vehicle cyberattacks. .

Understanding the relationship between the drivers' survey data and the responses to vehicle cyberattacks.

Investigating the drivers' cautionary behavior when facing vehicle cyberattacks.

Assessing the impact of level 2 automation and ADHD symptomatology on young drivers' attention maintenance.

Advancing the understanding of how safety on U.S. roads changed during the pandemic the number of fatal crashes during May through December 2020 to what would have been expected had the pandemic not occurred and pre-pandemic trends continued.