Erica Weng

I received my PhD from the Robotics Institute, part of the School of Computer Science at Carnegie Mellon University, where I worked on human trajectory forecasting with Professor Kris Kitani.

My research centers on perception and prediction for autonomous robots, including autonomous vehicles and social robotics.

I received my BS and MEng in Course 6 (Computer Science) from MIT, where I was a research assistant for Leslie Kaelbling in the Learning and Intelligent Systems group.

My research was supported by the Ford Foundation Fellowship Program (2022-2025).

Email  /  GitHub  /  Google Scholar  /  LinkedIn

Research

A Layered Foundation for Reliable Trajectory Forecasting: Data, Evaluation, and Methods

This thesis argues that reliable trajectory forecasting requires treating data curation, evaluation design, and modeling as co-equal engineering challenges, organized as a layered stack where each layer depends on the soundness of those below it.

Humanity's Last Exam

A multi-modal benchmark of 2,500 expert-level questions across dozens of subjects at the frontier of human knowledge. (Contributed 1 question.)

Evaluating a VR System for Collecting Safety-Critical Vehicle-Pedestrian Interactions

A user study evaluation of a virtual reality (VR) system for collecting safety-critical vehicle-pedestrian interaction data shows high perception of realism and user immersion as well as high similarity of data collected in the VR system to that collected in the real-world.

JaywalkerVR: A VR System for Collecting Safety-Critical Pedestrian-Vehicle Interactions

A virtual reality (VR) system based on the HTC VIVE VR headset system for collecting safety-critical vehicle-pedestrian interaction data that is difficult to collect and rare to find in the real world.

Joint Metrics Matter: A Better Standard for Trajectory Forecasting

A comprehensive evaluation of baselines with respect to multi-agent multimodal metrics (Joint ADE / FDE) and a general optimization method for improving Joint ADE / FDE.

Fast Neural Relational Inference with Modular Meta-Learning

An application of the Modular Meta-Learning algorithm to the problem of Neural Relational Inference outperforms previous non-modular deep learning methods.

Thanks to Leo for his cool website template.