IRAS

The Intelligent and Reliable Autonomous Systems (IRAS) research group focuses on the foundations and applications of sequential decision making, which enables autonomous systems to make complex decisions, while coping with uncertainty, limited information and resources. Our group is particularly interested in ensuring that the single and multi-agent systems are safe, reliable, and unbiased, when operating in fully and partially observable environments. Our methods draw insights from automated planning, reinforcement learning, robotics, decision theory, machine learning, and human factors. 

Our current research projects spans four broad themes: 

  1. Building safe and reliable autonomous systems: How can we support the design and deployment of safe and reliable autonomous systems in the open world?
  2. Sim-to-real transfer: How to develop approaches for efficient sim-to-real policy transfer?
  3. Human-centered AI: How can we design AI systems that are explainable, equitable, and mindful of human preferences and shortcomings?
  4. Planning under uncertainty: How can we efficiently express and solve problems involving planning under uncertainty and resource constraints, in single and multi-agent settings?  

Recent News

  • March 2026: IRAS member Alina Hyk recognized with Honors College Joe Hendricks Scholarship for Academic Excellence! Congratulations, Alina!
  • March 2026: Our paper on learning transferable latent user preferences has been accepted to the ICLR workshop on “From Human Cognition to AI Reasoning: Models, Methods, and Applications"! 
  • January 2026: Our paper on reward learning from diverse forms of human feedback is now published in the Frontiers of AI and Robotics! 
  • August 2025: Our WOFOSTGym paper was recognized with an Outstanding Applications Paper Award at RLC 2025! 
  • June 2025: Sandhya Saisubramanian gave an invited talk at the RSS Workshop on Multi-Objective Optimization and Planning in Robotics
  • May 2025: Our paper introducing WOFOSTGym crop simulator for RL has been accepted to RLC 2025!
  • January 2025: Paper on using credit estimation for mitigating side effects in multi-agent systems accepted to ICRA 2025!