Parv Kapoor

Ph.D. Student in Software Engineering
Software and Societal Systems Department
School of Computer Science
Carnegie Mellon University

parvk [at] cs [dot] cmu [dot] edu
CV / Google Scholar / Linkedin

Update: I am defending my thesis in Summer 2026 and am an RSS Pioneer 2026!

Hi! I am a Ph.D. in Software Engineering student at Carnegie Mellon University advised by Dr. Eunsuk Kang and Dr. Sebastian Scherer. During my Ph.D., I have interned at Microsoft Research, where I worked with Dr. John Langford on scaling long-horizon reasoning in large language models using Matryoshka representations. I was also a researcher with General Robotics working with Dr. Ashish Kapoor on developing safe robot foundation models.

I am interested in safe and robust robot learning and my work has been focused on:

1. Safe deep learning-based robotics
   • Vision-based control barrier functions for Aerial Robotics (RSS 25)
   • Using pretrained embeddings based behavior specifications for planning with world models (Under submission)
   • Efficient differentiable logic networks for trajectory optimization and diffusion policies (RA-L, ICRA 2026)
   • Embedding logical rules into autoregressive pre-trained transformers for trajectory planning (RSS SA 24)
   • Embedding general aviation rules into learning from demonstration policies using MCTS (ICRA 23)
2. Bridging the sim-to-real gap for deploying learned policies
   • Constrained Decoding for Robot Foundation Models (ICML 26)
   • Enhancing policy interpretability through trajectory analysis (ICSE 25)
   • Identifying failure cases of safe reinforcement policies using simulation guided search (FM 24)
   • Requirement decomposition for integrated task and motion planning (NFM 24)

Beyond my core research, I have also developed a custom deep learning library with implicit differentiation layers from scratch, conducted pilot user studies for trustworthy human-robot interaction, and contributed to the simulation and the foundation model workstreams for the General Robotics’ GRID platform.

I am currently a part of AirLab and SoDA. Prior to joining CMU, I worked with Dr. Jyotirmoy Deshmukh’s CPS VIDA lab at University of Southern California. I have also researched with Dr. Thao Dang at the Université Grenoble Alpes and Dr. David Marshall at the Cardiff University.

Reviewer: RSS 2026, RSS Pioneers 2026, ICLR 2026, IASEAI 2026, L4DC 2025, RSS 2025, CoRL 2026

Teaching Assistant: 10-623 Generative AI, 17-723 Designing Large Scale Systems at CMU