Nir Pechuk

Pittsburgh, PA

Hi, I’m Nir! I’m a computer science undergraduate at Carnegie Mellon University, interested in reinforcement learning, robot learning, and building systems that make a difference in people’s lives.

I recently joined the AI & Reinforcement Learning (AIRe) Lab under Prof. Aviral Kumar. I also work with Prof. Zackory Erickson in the Robot Caregiving and Human Interaction (RCHI) Lab, where we turn text prompts into simulated human–robot interactions and policies that run on real robots.

This fall, I’m a 15-213: Introduction to Computer Systems TA, after TAing 15-122 in the spring. I love helping students build things from the ground up, from data structures to memory allocators and web proxies. I’m also the Research Chair of ACM@CMU.

I’ve built infrastructure at Palantir, machine learning and database systems at Firebolt, and developer tools at SDF Labs. Before CMU, I co-founded Extentek and invented a patented device to help visually impaired people pour independently.

Outside research, I enjoy hackathons and projects that connect software with the physical world. Recent builds include an autonomous bioreactor, a tool for understanding local elections, and an agentic fact-checking browser extension.

Email GitHub LinkedIn CV

selected publications

GenPHRI: Agentic Generative Simulation for Physical Human-Robot Interaction

GenPHRI turns a single prompt into a simulated human, environment, robot, and motion, supporting zero-shot prompt2sim2real transfer for real-world caregiving tasks.

Junxiang Wang^*, Nir Pechuk^*, Xinwen Xu^*, Tiancheng Wu, Nayon Kim, Julian Millan, and Zackory Erickson

* Equal contribution

Preprint, 2026

arXiv Bib Video Website
```
@article{wang2026genphri,
  title = {GenPHRI: Agentic Generative Simulation for Physical Human-Robot Interaction},
  author = {Wang, Junxiang and Pechuk, Nir and Xu, Xinwen and Wu, Tiancheng and Kim, Nayon and Millan, Julian and Erickson, Zackory},
  journal = {Preprint},
  year = {2026},
}
```
What Should Embodied Agents Remember? Benchmarking and Evaluating Memory Modality for Embodied Agents

MeMoREval benchmarks how memory modality affects embodied agents across 53 tasks, measuring task success and the inference costs of building and using memory.

Junxiang Wang, Yuchen Liu, Yi Wu, Guofei Chen, Nir Pechuk, Tao Chen, Zackory Erickson, Yuzhe Qin, and Tianyu Li

Preprint, 2026

arXiv Bib GitHub Website
```
@article{wang2026memoreval,
  title = {What Should Embodied Agents Remember? Benchmarking and Evaluating Memory Modality for Embodied Agents},
  author = {Wang, Junxiang and Liu, Yuchen and Wu, Yi and Chen, Guofei and Pechuk, Nir and Chen, Tao and Erickson, Zackory and Qin, Yuzhe and Li, Tianyu},
  journal = {Preprint},
  year = {2026},
}
```

Spill-O-Not attached to a blue mug, from Figure 3(a) of the paper

Spill-O-Not: A contactless material level indicator for visually impaired users

The Galina / Spill-O-Not project uses contactless LiDAR time-of-flight sensing and audio feedback to help visually impaired users avoid overpouring.

Maxwell Mamishev, Nir Pechuk, Rayhan Shaik, and Braden Miller

Journal of Physics: Conference Series, 2026

DOI Bib Paper Website

@article{mamishev2026spill,
  title = {Spill-O-Not: A contactless material level indicator for visually impaired users},
  author = {Mamishev, Maxwell and Pechuk, Nir and Shaik, Rayhan and Miller, Braden},
  journal = {Journal of Physics: Conference Series},
  volume = {3216},
  number = {1},
  pages = {012016},
  year = {2026},
  doi = {10.1088/1742-6596/3216/1/012016},
}

Figure 2 of U.S. Patent 10,830,626 B2, showing a contactless level-sensing device attached to the rim of a container

Container filling or emptying guidance device

A contactless sensing device monitors changing material levels and provides feedback to guide container filling or emptying and help prevent spills.

Nir Pechuk, Ron Pechuk, and Maxwell A. Mamishev

U.S. Patent No. 10,830,626 B2, Nov 2020

Bib PDF Website

@misc{pechuk2020container,
  title = {Container filling or emptying guidance device},
  author = {Pechuk, Nir and Pechuk, Ron and Mamishev, Maxwell A.},
  howpublished = {U.S. Patent No. 10,830,626 B2},
  year = {2020},
  month = nov,
  day = {10},
  url = {https://patents.google.com/patent/US10830626B2/en},
}