Location: Zoom
Add to Calendar 2023-03-21T11:00:00 2023-03-21T13:00:00 America/New_York PhD Proposal: Shili Sheng All Invited Committee Members: Haiying Shen (Chair) - ESE, CS, ECE Lu Feng (Advisor) - ESE, CS Tariq Iqbal - ESE, CS Homa Alemzadeh - ECE, CS Seongkook Heo - CS For Zoom information, please send an email to ese-programs@virginia.edu. Title:  Enhancing Safety and Trustworthiness for Autonomous Agents under Uncertain Effects Abstract:  Zoom

All Invited

Committee Members:

  • Haiying Shen (Chair) - ESE, CS, ECE
  • Lu Feng (Advisor) - ESE, CS
  • Tariq Iqbal - ESE, CS
  • Homa Alemzadeh - ECE, CS
  • Seongkook Heo - CS

For Zoom information, please send an email to ese-programs@virginia.edu.

Title:  Enhancing Safety and Trustworthiness for Autonomous Agents under Uncertain Effects

Abstract: 

Safety and trustworthiness are essential for autonomous agents to work successfully with humans. Ensuring their ability to fulfill their objectives as desired is one approach to enhance their safety and trustworthiness. Hence, it is crucial to design effective policies to guide the agents. However, designing policies for agents under uncertain effects can be difficult. Many agents that work with humans are unable to directly measure the states of human minds, which are often uncertain and dynamically evolving. Moreover, the agents can also be flawed due to noisy sensors and imperfect controllers. We aim to study these challenges in enhancing safety and trustworthiness for autonomous agents by designing effective policies based on Partially Observable Markov Decision Process (POMDPs).  We first studied the policy planning for automated vehicles while accounting for human trust and intention. Furthermore, we investigated a safe online policy planning method integrated with shielding techniques to ensure that the agent can reach its destination and avoid unsafe states. We propose to extend our prior work to synthesize permissive policies while accounting for uncertainties from both humans and autonomous agents.