Harsh Goel

I am a Ph.D. student at The University of Texas at Austin. My research develops video agents that search for relevant information, reason about events and human intent, and coordinate with people and other agents. Recent work includes long-video question answering, neuro-symbolic video reasoning, and multi-agent reinforcement learning.

I was previously advised by Dr. Guillaume Sartoretti and Dr. Marcelo H. Ang Jr. at the National University of Singapore, and Dr. Vijay Kumar at the University of Pennsylvania. Outside the lab, I enjoy tennis, AI podcasts, and playing the keyboard.

Download my resume

news

Jun 18, 2026	Two papers were accepted at ECCV 2026: VSeek and CrossView.
Jan 20, 2026	NeuS-QA: Grounding Long-Form Video Understanding in Temporal Logic and Neuro-Symbolic Reasoning was accepted at AAAI 2026.
Aug 6, 2025	Distributed Upload and Active Labeling for Resource-Constrained Fleet Learning was accepted at CoRL 2025.
Jun 2025	CoordLight: Learning Decentralized Coordination for Network-Wide Traffic Signal Control was published in IEEE Transactions on Intelligent Transportation Systems.
May 4, 2025	R3DM: Enabling Role Discovery and Diversity Through Dynamics Models in Multi-Agent Reinforcement Learning was accepted at ICML 2025.
Mar 15, 2025	NeuS-V: Neuro-Symbolic Evaluation of Text-to-Video Models using Formal Verification was accepted at CVPR 2025.
Jul 4, 2024	Towards Neuro-Symbolic Video Understanding was accepted at ECCV 2024.
May 2, 2024	Improving End-To-End Autonomous Driving with Synthetic Data from Latent Diffusion Models was accepted at the CVPR Vision and Language for Autonomous Driving and Robotics Workshop.
Oct 16, 2023	Invited reviewer for ICRA 2024
Aug 8, 2023	My MS Thesis talk on Hierarchical Reinforcement Learning for Aerial Vehicles is availble at https://www.youtube.com/watch?v=DM9q31wrbBI&t=338s.
May 10, 2023	Accepted a PhD offer with a Cockrell School of Engineering fellowship from the University of Texas at Austin
Mar 2, 2023	Opensourced code for SocialLight: Distributed Cooperation Learning for network wide traffic signal control accepted to AAMAS 2023.
Jan 13, 2023	Currently looking for a PhD/Research Engineering position.
Jan 6, 2023	Paper on Cooperative traffic signal control accepted to AAMAS 2023.
Oct 28, 2022	Glad to annouce that our paper on agile informative path planning for quadrotors with reinforcement learning was presented at the AI2R Worshop at IROS 2022 in Kyoto Japan.

publications

2026

ECCV

Incentivizing Vision Language Models to Search for Long Video Question Answering

Harsh Goel^*, S P Sharan^*, Sahil Shah, and 4 more authors

* Equal contribution

In European Conference on Computer Vision (ECCV), 2026

arXiv Website
ECCV

CrossView: Can Vision-Language Models Reason Across Cameras?

Sahil Shah^*, S P Sharan^*, Harsh Goel^*, and 4 more authors

* Equal contribution

In European Conference on Computer Vision (ECCV), 2026

arXiv Website
AAAI

NeuS-QA: Grounding Long-Form Video Understanding in Temporal Logic and Neuro-Symbolic Reasoning

Sahil Shah, S P Sharan^*, Harsh Goel^*, and 5 more authors

* Equal contribution

Proceedings of the AAAI Conference on Artificial Intelligence, 2026

arXiv HTML
arXiv

Bayesian Partner Modelling enables Adaptive Replanning for LLM Coordination

Harsh Goel, Aditya Sai Ellendula, Vaishnav Tadiparthi, and 3 more authors

2026

arXiv preprint

arXiv
arXiv

S^3-R1: Learning to Retrieve and Answer Step-by-Step with Synthetic Data

Harsh Goel, Akhil Udathu, Susmija Jabbireddy, and 2 more authors

2026

arXiv preprint

arXiv

2025

CVPR

Neuro-Symbolic Evaluation of Text-to-Video Models using Formal Verification

S P Sharan, Minkyu Choi, Sahil Shah, and 3 more authors

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

HTML PDF
ICML

R3DM: Enabling Role Discovery and Diversity Through Dynamics Models in Multi-Agent Reinforcement Learning

Harsh Goel, Mohammad Omama, Behdad Chalaki, and 3 more authors

In Proceedings of the 42nd International Conference on Machine Learning (ICML), 2025

HTML PDF Code
CoRL

Distributed Upload and Active Labeling for Resource-Constrained Fleet Learning

Oguzhan Akcin, Harsh Goel, Ruihan Zhao, and 1 more author

In Proceedings of the 9th Conference on Robot Learning (CoRL), 2025

HTML PDF
T-ITS

CoordLight: Learning Decentralized Coordination for Network-Wide Traffic Signal Control

Yifeng Zhang, Harsh Goel, Peizhuo Li, and 3 more authors

IEEE Transactions on Intelligent Transportation Systems, 2025

PDF Code

2024

CVPRW

Improving End-To-End Autonomous Driving with Synthetic Data from Latent Diffusion Models

Harsh Goel, Sai Shankar Narasimhan, and Sandeep P. Chinchali

In First Vision and Language for Autonomous Driving and Robotics Workshop, 2024

HTML PDF
ECCV

Towards Neuro-Symbolic Video Understanding

Minkyu Choi, Harsh Goel^*, Mohammad Omama^*, and 3 more authors

* Equal contribution

In European Conference on Computer Vision (ECCV), 2024

arXiv HTML

2023

AAMAS

SocialLight: Distributed Cooperation Learning towards Network-Wide Traffic Signal Control

Harsh Goel, Yifeng Zhang, Mehul Damani, and 1 more author

In Proceedings of the 22nd International Conference on Autonomous Agents and Multiagent Systems, 2023

Abs PDF

Many recent works have turned to multi-agent reinforcement learn- ing (MARL) for adaptive traffic signal control to optimize the travel time of vehicles over large urban networks. However, achieving ef- fective and scalable cooperation among junctions (agents) remains an open challenge, as existing methods often rely on extensive, non- generalizable reward shaping or on non-scalable centralized learn- ing. To address these problems, we propose a new MARL method for traffic signal control, SocialLight, which learns cooperative traffic control policies by distributedly estimating the individual marginal contribution of agents on their local neighborhood. SocialLight re- lies on the Asynchronous Actor Critic (A3C) framework, and makes learning scalable by learning a locally-centralized critic conditioned over the states and actions of neighboring agents, used by agents to estimate individual contributions by counterfactual reasoning. We further introduce important modifications to the advantage calculation that help stabilize policy updates. These modifications decouple the impact of the neighbors’ actions on the computed advantages, thereby reducing the variance in the gradient updates. We benchmark our trained network against state-of-the-art traf- fic signal control methods on standard benchmarks in two traffic simulators, SUMO and CityFlow. Our results show that SocialLight exhibits improved scalability to larger road networks and better performance across usual traffic metrics.

2022

IROS - AI2R

Reinforcement Learning for Agile Active Target Sensing with a UAV

Harsh Goel, Laura Jarin Lipschitz, Saurav Agarwal, and 2 more authors

2022

arXiv