Karan Muvvala

Karan Muvvala

Graduate Research Assistant

University of Colorado Boulder

Biography

Hey, I am Karan, a Third-year Ph.D. student in the Aerospace Engineering Dept. at CU. I am currently working as a Research Assistant at Aria Systems advised by Dr. Morteza Lahijanian. At ARIA Systems group I work on developing theories and algorithms that enable robots to operate safely, robustly, and efficiently in unknown and dynamic environments. Currently, I am working on

  • Verifying the safety of complex autonomous systems modeled as neural networks and certifying their behaviors using Barrier Methods for Safe Autonomy.
  • And Synthesizing high-level strategies for a robotic manipulator operating in the presence of a human using game-theoretic approaches to enable more human-like behavior for robots.

My current research interest includes Formal Methods, Verification and Planning for Autonomous systems, Game Theory, and Control Theory.

Please reach out to if you have any questions in general at - karan.muvvala@colorado.edu. Thank you!

Checkout our lab’s github repo to see what I have been upto!

I am actively looking for internship opportunities for Summer 24. Specifically, I would like to work on research topics related to

  • Human-centered autonomy
    • Interaction-aware efficient decision making for safety-critical systems (ICRA22; IROS23; NIPS22)
  • Assured autonomy
    • Safety assurances with machine learning models in the loop (NIPS22)

I am always open to new ideas and application fields. Please reach out to me at karan.muvvala@colorado.edu if you have any questions. :)

Interests
  • Formal Methods
  • Neural Network Verification
  • Game Thoery
  • Control Theory
Education

  • PhD in Aerospace Engineering Sciences - Autonomous Systems Track, Present

    University of Colorado Boulder

  • MS in Mechanical Engineering - Robotics and Systems Design Track, 2021

    University of Colorado Boulder

  • BE in Mechanical Engineering, 2018

    University of Mumbai

News

One paper was accepted at the NeurIPS 2022 conference. Click here for more info.
TL;DR: In this work, we introduce a method of safety certification and control for neural network dynamic systems via stochastic barrier functions.
Karan Muvvala
Last updated on Oct 2, 2022
PDF Code Video arXiv
One paper was accepted at the NeurIPS 2022 conference. Click here for more info.
I was selected for the IEEE MGA Volunteer Leadership Training (VoLT) Program!
This program aims at traning new leaders who have a desire to continue growing as IEEE volunteers and lead their local units. This program will help me build crucial leadership and mentoring skills that will help me in my mentoring and outreach activities.
Karan Muvvala
Last updated on Oct 2, 2022
VoLT Program
One paper accepted for presentation at IEEE Aerospace Conference, 2023. Click Here for details!
TL;DR: In this work, we introduce a Preference based correct-by-construction planning framework (REASON) for efficient exploration of icy moons like Europa for Autonomous Landers.
Last updated on Oct 2, 2022
An extended abstract got accepted at RSS 2022 Workshop on Risk-aware Decision Making
Last updated on Oct 2, 2022
PDF Slides
Selected for RSS Inclusion 2022 Fellowship
Last updated on Oct 2, 2022
Fellowship
See all posts

Recent Publications

Efficient Symbolic Approaches for Quantitative Reactive Synthesis with Finite Tasks (IROS 23)
In this work, we introduce a method of symbolic value iteration for quantitative tow-player games with reachability objectives and with resource constraints.
Karan Muvvala, Morteza Lahijanian
PDF Cite Code Video Experiments
Efficient Symbolic Approaches for Quantitative Reactive Synthesis with Finite Tasks (IROS 23)
Stochastic Games for Interactive Manipulation Domains
In this work, we explore stochastic games as an abstraction for modeling human-robot interaction with stochastic action outcomes and irrational agents. We also improve scalabiltiy of exisitng tool PRISM-games.
Karan Muvvala, Andrew Wells, Morteza Lahijanian, Lydia Kavraki, Moshe Vardi
Cite
Stochastic Games for Interactive Manipulation Domains
REASON-RECOURSE Software for Science Operations of Autonomous Robotic Landers (accepted IEEE AeroConf 23)
In this work, we introduce a Preference based correct-by-construction planning framework (REASON-RECOURSE) for efficient exploration of icy moons likes Europa.
Jay McMahon, Nisar Ahmed, Morteza Lahijanian, Peter Amorese, Taralicin Deka, Karan Muvvala, Trevor Slack, Shohei Wakayama, Kian Shakerin
PDF Cite
REASON-RECOURSE Software for Science Operations of Autonomous Robotic Landers (accepted IEEE AeroConf 23)
Safety Guarantees for Neural Network Dynamic Systems via Stochastic Barrier Functions (accepted NeurIPS 22)
In this work, we introduce a method of safety certification and control for neural network dynamic systems via stochastic barrier functions.
Rayan Mazouz, Karan Muvvala, Akash Ratheesh, Luca Laurenti, Morteza Lahijanian
PDF Cite Code Video OpenReview
Safety Guarantees for Neural Network Dynamic Systems via Stochastic Barrier Functions (accepted NeurIPS 22)
Let's Collaborate: Regret-based Reactive Synthesis for Robotic Manipulation
Synthesizing human-like behavior for robots by reasoning about the human’s action to synthesize smarter interactions.
Karan Muvvala, Peter Amorese, Morteza Lahijanian
Cite Code Poster Slides Video ArXiv
Let's Collaborate: Regret-based Reactive Synthesis for Robotic Manipulation
See all publications

Contact

The easiest way to get in touch with me is via email. Feel free to reach out to me if you are have any questions about my research, academia or robotics in general.