I am a final-year M.S. student in Computer Science at Brown University, advised by George Konidaris, Daniel Ritchie, and Thomas Serre. My research focuses on world models, generative modeling, and embodied intelligence, with an emphasis on learning representations of the physical world for planning and decision-making.

I'm currently visiting the Kempner Institute at Harvard University, hosted by Qianqian Wang. Previously, I conducted research at the National University of Singapore (NUS). I earned my B.S. in Computer Science from Fort Hays State University (FHSU).

I've been fortunate to collaborate with wonderful researchers and mentors, including Victor Boutin, Sergio Orozco, Jorge Chang, and many others.

I usually go by Lyfey. Coincidentally, my initials are “CV”—quite fitting for someone working on computer vision!

I'm seeking PhD opportunities for Fall 2027 in machine learning, computer vision, and robotics. Please feel free to reach out!

News

  • Oct 09, 2026. Excited to head to Texas next month for the CoRL 2026 Workshop on Do Robots Need World Models? Looking forward to the discussions by Vincent Sitzmann, Chelsea Finn, and Yunzhu Li. Howdy Texas! 🤠
  • Oct 03, 2026. Visited Princeton for the Northeast Robotics Colloquium. Thanks to George Konidaris for supporting the trip!
  • Sep 09, 2026. Began visiting the Kempner Institute at Harvard University during the final year of my M.S. Thanks to Brown and Qianqian Wang for the opportunity!
  • Jul 06, 2026. Started my master's thesis on Learning World Models for Robotics with George Konidaris.
  • May 26, 2026. Received NSF-supported funding for summer research. Special thanks to Thomas Serre!
  • Apr 20, 2026. Presented Geodesic Interpolation with Diffusion and Flow Models at Brown's Visual Computing Seminar. Thanks to Daniel Ritchie and James Tompkin for hosting!
  • Apr 12, 2026. Visited MIT for the New England Symposium on Graphics.
  • Sep 04, 2025. Started my M.S. in Computer Science at Brown University.

Publications

TAP-MPC: Transferable Action Priors for Model Predictive Control with World Models

TAP-MPC: Transferable Action Priors for Model Predictive Control with World Models

Sergio Orozco, Chealyfey Vutha, Nandika Auluck, Anthony Ramirez-Roca, and George Konidaris

We train an action prior with a VQ-VAE, roll out its samples in a world model, and refine the best rollout with MPC, making planning tractable for robotic manipulation.

Poster

Not Too Generative, Not Too Discriminative: The Human Alignment Sweet Spot

Computational Alignment of EBMs with Humans

Chealyfey Vutha, Jorge Chang, Bastien Le Lan, Victor Boutin, and Thomas Serre

Under Submission

We use Joint Energy-Based Models (JEMs) to interpolate between discriminative and generative training within a fixed architecture, and show that human alignment is maximized at intermediate points along this continuum rather than at either extreme.

PDF · Code · Project Page

Geodesic Interpolation for Diffusion and Flow Models

Geodesic Interpolation for Diffusion and Flow Models

Chealyfey Vutha, Zheyu Shi, Yuqiao Guan, and Daniel Ritchie

Under Submission

We study interpolation in diffusion and flow models by following smooth paths, which enables smoother and more semantically consistent transitions between generated images.

PDF · Code · Project Page

Experience

  • Jan 2025 – Aug 2025— AI Engineer, Ministry of Post and Telecommunications (MPTC).

    Led the development of the speech-to-speech translation feature for TranslateKH, Cambodia's national translation application.

  • Jul 2023 – Dec 2024— Student Researcher, National University of Singapore (NUS).

    Co-developed a research proposal for Heritage Plate, a storytelling recipe application, securing a $5,000 grant from the Asian Undergraduate Symposium at NUS.

Service & Teaching

Teaching

Academic Service