Hi, I'm Mehrnaz.
PhD candidate @ Cornell University
Working on Collaborative Autonomy
My research aims to enable machines to learn how to coordinate with humans as a team towards shared goals. I believe learning effective cooperative behaviors can ultimately enable reliable autonomous systems. I build on emerging techniques in Multi-Agent Machine Learning and apply them to new training environments that are informed by real-world application scenarios involving drones.
In the News
ACHIEVEMENTS AND AWARDS
Partnerships for AI-enabled Traffic Management for Advanced Air Mobility
Led industry engagement for my NASA-funded project and secured more than 10 key partnerships with leading stakeholders and industry players for the project’s execution. Raised grant pool by 15% through additional fundraising
2024
NASA University Student Research Challenge – Grant: $80k
Awarded to student research projects with novel approaches to solving some of the biggest technical challenges facing aviation as identified by NASA’s Aeronautics Research Mission Directorate
2024
Nvidia-NASA Hackathon
Selected as part of an exclusive cohort of teams working on large-scale NASA-funded projects. Collaborated with Nvidia mentors in a 1-month program to optimize AI training compute and accelerate large-scale simulation-based learning
2024
Global Advanced Air Mobility Academic Competition Finalist
2024
NSF Innovation Corps National Award – Grant: $50k
Awarded to top researchers in science and engineering fields with promising lab inventions
2023
Cornell Engineering Commercialization Fellowship
Awarded to three Cornell Engineering PhD candidates with research-based impactful technology innovations
2023
NSF Spirit of I-Corps Award
Awarded for demonstrating excellence in leadership and execution during the national program
2023
ACM Best Paper Award
2022
Best ECE Undergraduate Thesis Project for Fundamental Design and Innovation
2020
Best Computer Eng. Undergraduate Thesis Project
2020
(4 x) Best Undergraduate Thesis Project by Industry
Awarded by 4 different commercial companies
2020
PROFESSIONAL EXPERIENCE
Lead Researcher on Human-Drone Systems
Cornell University
Collaborative Technologies Lab
- Currently working on implementing a new multi-agent machine learning algorithm for learning cooperative policies from humans in partially observable multi-agent simulation environments
- Implemented a suite of multi-agent human-in-the-loop cooperative simulation training environments informed by real-world search missions collected from more than 200 search and rescue drone operations
- Created an optimized training pipeline for on-policy MARL baseline finetuning in high-fidelity simulation environments
- Conducted studies on a new system for human-machine teaming data collection using a multi-person human-drone platform [paper submission underway]
2021 - Present
PI/Team Lead — AI-enabled Traffic Management for Advanced Air Mobility
NASA
-
Building on my awarded NASA proposal and collaborating with attracted industry partners to implement a new traffic management platform for urban air mobility utilizing AI. This project can be found at projectorion.info
-
Building on my PhD research work to develop and test new cooperative machine learning models for advanced urban air mobility scenarios
-
Implementing end-to-end data engine for training and testing multi-agent learning baselines in new simulation configurations that represent challenges faced in urban air mobility scenarios
-
Implementing a city-scale high-fidelity simulation environment for performant aerial data collection as well as software and hardware in-the-loop testing
-
Leading a multi-disciplinary team of undergraduate and master students across CS, IS, and ECE
2024-Present
Nvidia-NASA Hackathon
-
Pretrained and optimized two state-of-the-art navigation transformer models on DGX cloud. By leveraging accelerated computing, we achieved a 55% improvement in training performance and a 40% increase in memory efficiency
-
Developed a more compact navigation model to accelerate deployment in scenarios where balancing performance and generalizability is crucial
-
Created reusable pipelines for mixed precision, kernel fusion and distributed training for our ongoing work in the project to increase compute utilization
-
Results were shared as part of Open Accelerated Computing (OAC) Summit’24 under dedicated session: Accelerating AI for Autonomous Navigation: Optimizing Navigation Transformers for Large-Scale Use Cases
NSF I-Corps – National Program – Entrepreneurial Lead
-
Led customer discovery and market validation efforts for a simulation platform I implemented for validating the reliability of autonomy stacks in robot vehicles
-
Engaged with over 180 industry stakeholders (across markets of ground robotics, AV, and aerospace) to identify commercialization pathways
June-Aug 2023
Microsoft — Autonomous Systems Group Intern
-
Designed, developed and studied a new procedural generative aerial synthetic training data augmentation framework that increases end data variants by 75% and contributes to increased generalizability of trained models
-
Designed and implemented an adaptive Domain Randomization approach for type-agnostic realistic scene augmentation to address sim-to-real
-
The data augmentation pipeline was directly deployed to the end solution as part of the AirSim software suite
-
Proposed a new iterative data collection optimization approach for efficiently generating synthetic aerial datasets to meet a performance target
-
Ran experiments on representation learning methods for localization towards improving vision-based drone navigation using multi-modal synthetic data
June-Aug 2022
PROJECTS
Advanced simulation system for validating autonomous robots
Cornell University
- Implemented an advanced modular simulation system to validate the autonomy behaviors in autonomous robots. This system provides scalable testbeds for robot autonomy through a modular data- driven environment
- Implemented a novel approach in multi-modal data collection, through a performant rendering mechanism that enables the simulator to generate large-scale test data to validate autonomy behaviors for single or multi-robot scenarios thus mitigating risk of deployment and accelerating test and validation
- This system has won 2 major awards including an NSF grant and is actively being used for research
2021 - 2023
Integrating quadcopter drones to ad-hoc operations during disaster response
B.Sc. Thesis – Department of Electrical and Computer Engineering
-
Studied first responders’ collaboration during natural disasters and designed a platform prototype using quadcopter drones, advanced cloud computing, machine learning and data visualization techniques to support and facilitate drone operations in distributed teams for effective response.
-
This project won 6 awards from industry and academia in total.
2018-2020
SELECTED PUBLICATIONS AND PATENTS
Sabet, M., 2024. Intelligent Testbeds for Aerial Autonomy Assurance in Cooperative Airspace via Large Language Models, under review in Global AAM Academic Paper Competition for submission in the Drone Systems and Applications journal
Sabet, M., Palanisamy, P., & Mishra, S. (2023). Scalable modular synthetic data generation for advancing aerial autonomy. Robotics and Autonomous Systems, 166, 104464.
-
This paper addresses generalizability and scalability challenges in aerial synthetic data generation for sim-to-real by:
-
Introducing ASDA, procedural generative aerial synthetic data augmentation framework for generating diverse aerial synthetic datasets.
-
Adaptive scene augmentation through a multi-layer domain randomization approach guided by scene graph information.
-
Increasing aerial synthetic data generation efficiency by enabling flexible control through a unified prompt-based interface.
-
Our approach results in 75% more variants in the end data and increases data generation efficiency by 50%.
-
Iterative data collection optimization method based on ASDA’s workflow to meet a performance target.
Sabet, M., Orand, M., & W. McDonald, D. (2021). Designing Telepresence Drones to Support Synchronous, Mid-air Remote Collaboration: An Exploratory Study. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems.
-
This paper describes the design and development of a novel telepresence platform using drones that supports collaboration and mid-air interaction among multiple remote and onsite users through three design iterations and evaluations
Sabet, M. 2024. Neural engine for training distributed neural networks in simulated environments. Submitted to The United States Patent and Trademark Office.
Sabet, M. 2024. System and methods for generating synthetic validation data for adaptive autonomous machines. Submitted to The United States Patent and Trademark Office.
Sabet, M. 2024. Multi-agent human-in-the-loop simulation system for human-autonomy teaming training and evaluation. Submitted to The United States Patent and Trademark Office.
PROFESSIONAL ACTIVITIES
Project Director, Shaping Autonomy
-
Conducted more than 100 interviews with industry experts in Robotics and Drones sector to identify real-world technical gaps and challenges inspiring applied cross-institutional projects that address critical gaps and empower the research and engineering community
-
Initiative backed by NSF and supported by three major partnerships with leading industry associations (AUVSI, Women&Drones, P3 Tech Consulting)
2023-2024
Technical Committee Member, Human-Machine Teaming, AIAA
2024-present
Friend of the Technical Committee, Intelligent Systems, AIAA
2024-present
Multi-Vehicle Control (m:N) Working Group member, NASA
2022-present
Associate member, Association for Uncrewed Vehicle Systems International (AUVSI)
2022-present
Speaker, Open Accelerated Computing (OAC) Summit’24, Accelerating AI for Autonomous Navigation:
Optimizing Navigation Transformers for Large-Scale Use Cases