Ethical Decision Making

TLDR: Enable autonomous systems to comply with ethical theories.

Ethically Compliant Autonomous Systems under Partial Observability
Qingyuan Lu, Justin Svegliato, Samer Nashed, Shlomo Zilberstein, Stuart Russell
International Conference on Robotics and Automation (ICRA)
— ICRA 2024
Ethically Compliant Planning within Moral Communities
Samer Nashed, Justin Svegliato, Shlomo Zilberstein
Conference on AI, Ethics, and Society (AIES)
— AIES 2021
Ethically Compliant Sequential Decision MakingDistinguished Paper Award
Justin Svegliato, Samer Nashed, Shlomo Zilberstein
AAAI Conference on Artificial Intelligence (AAAI)
— AAAI 2021
An Integrated Approach to Moral Autonomous Systems
Justin Svegliato, Samer Nashed, Shlomo Zilberstein
European Conference on Artificial Intelligence (ECAI)
— ECAI 2020

Adversarial Robustness

TLDR: Safeguard LLMs against adversarial attacks.

A StrongREJECT for Empty Jailbreaks
Alexandra Souly, Qingyuan Lu, Dillon Bowen, Tu Trinh, Elvis Hsieh, Sana Pandey, Pieter Abbeel, Justin Svegliato, Scott Emmons, Olivia Watkins, Sam Toyer
arXiv preprint arXiv:2402.10260
— arXiv 2024
Tensor Trust: Interpretable Prompt Injection Attacks from an Online GameSpotlight Talk
Sam Toyer, Olivia Watkins, Ethan Mendes, Justin Svegliato, Luke Bailey, Tiffany Wang, Isaac Ong, Karim Elmaaroufi, Pieter Abbeel, Trevor Darrell, Alan Ritter, Stuart Russell
International Conference on Learning Representations (ICLR)
— ICLR 2024

State Abstractions

TLDR: Solve MDPs using partial state abstractions.

Selecting the Partial State Abstractions of MDPs: A Metareasoning Approach with Deep RL
Samer Nashed*, Justin Svegliato*, Abhinav Bhatia, Shlomo Zilberstein, Stuart Russell
International Conference on Intelligent Robots and Systems (IROS)
— IROS 2022
Solving Markov Decision Processes with Partial State Abstractions
Samer Nashed*, Justin Svegliato*, Matteo Brucato, Connor Basich, Shlomo Zilberstein
International Conference on Robotics and Automation (ICRA)
— ICRA 2021

Exception Recovery

TLDR: Allow autonomous systems to recover from exceptions.

Introspective Autonomous Vehicle Operational Management
Justin Svegliato, Kyle Wray, Stefan Witwicki, Shlomo Zilberstein
US Patent 10,649,453
— US 2020
Belief Space Metareasoning for Exception Recovery
Justin Svegliato, Kyle Wray, Stefan Witwicki, Joydeep Biswas, Shlomo Zilberstein
International Conference on Intelligent Robots and Systems (IROS)
— IROS 2019

Optimal Stopping

TLDR: Optimize the stopping point of anytime planners.

A Model‑Free Approach to Meta‑Level Control of Anytime Algorithms
Justin Svegliato, Prakhar Sharma, Shlomo Zilberstein
International Conference on Robotics and Automation (ICRA)
— ICRA 2020
Meta‑Level Control of Anytime Algorithms with Online Performance Prediction
Justin Svegliato, Kyle Wray, Shlomo Zilberstein
International Joint Conference on Artificial Intelligence (IJCAI)
— IJCAI 2018

Agent-Aware State Estimation

TLDR: Enable autonomous systems to discover their state by observing the behavior of other agents.

Agent‑Aware State Estimation for Autonomous Vehicles
Shane Parr*, Ishan Khatri*, Justin Svegliato, Shlomo Zilberstein
International Conference on Intelligent Robots and Systems (IROS)
— IROS 2021

Competence Awareness

TLDR: Allow autonomous systems to learn/improve their competence via human feedback.

Competence‑Aware Systems
Connor Basich, Justin Svegliato, Kyle Wray, Stefan Witwicki, Joydeep Biswas, Shlomo Zilberstein
Artificial Intelligence Journal (AIJ)
— AIJ 2022
Improving Competence via Iterative State Space Refinement
Connor Basich, Justin Svegliato, Allyson Beach, Kyle Wray, Stefan Witwicki, Shlomo Zilberstein
International Conference on Intelligent Robots and Systems (IROS)
— IROS 2021
Learning to Optimize Autonomy in Competence‑Aware Systems
Connor Basich, Justin Svegliato, Kyle Wray, Stefan Witwicki, Joydeep Biswas, Shlomo Zilberstein
International Conference on Autonomous Agents and Multiagent Systems (AAMAS)
— AAMAS 2020

Safe Decision Making

TLDR: Enable autonomous systems to be more safe.

Defining Deception in Decision Making
Marwa Abdulhai, Micah Carroll, Justin Svegliato, Anca Dragan, Sergey Levine
International Conference on Autonomous Agents and Multiagent Systems (AAMAS)
— AAMAS 2024
Fairness and Sequential Decision Making: Limits, Lessons, and Opportunities
Samer Nashed, Justin Svegliato, Su Lin Blodgett
arXiv preprint arXiv:2301.05753
— arXiv 2023
Metareasoning for Safe Decision Making in Autonomous Systems
Justin Svegliato, Connor Basich, Sandhya Saisubramanian, Shlomo Zilberstein
International Conference on Robotics and Automation (ICRA)
— ICRA 2022

Optimal Hyperparameter Tuning

TLDR: Optimize the hyperparameters of anytime planners.

Tuning the Hyperparameters of Anytime Planning: A Metareasoning Approach with Deep RL
Abhinav Bhatia, Justin Svegliato, Samer Nashed, Shlomo Zilberstein
International Conference on Planning and Scheduling (ICAPS)
— ICAPS 2022
On the Benefits of Randomly Adjusting Anytime Weighted A*
Abhinav Bhatia, Justin Svegliato, Shlomo Zilberstein
Symposium on Combinatorial Search (SoCS)
— SoCS 2021

Agent Architectures

TLDR: Formally compose the architecture of agents.

Formal Composition of Robotic Systems as Contract Programs
Mason Nakamura, Justin Svegliato, Samer Nashed, Shlomo Zilberstein, Stuart Russell
International Conference on Intelligent Robots and Systems (IROS)
— IROS 2023

Active Teacher Selection

TLDR: Actively select teachers during RLHF.

Active Teacher Selection for Reinforcement Learning from Human Feedback
Rachel Freedman, Justin Svegliato, Kyle Wray, Stuart Russell
arXiv preprint arXiv:2310.15288
— arXiv 2023