Kevin Zakka

I am a PhD student at the Robot Learning Lab (RLL) at UC Berkeley, where I am advised by Pieter Abbeel. I work on robot learning for humanoid robots, with a focus on whole-body control.

Previously, I completed a Master's ('21) in Computer Science at Stanford University, under the supervision of Jeannette Bohg.

News

• Apr 25: Playground has been accepted to RSS 2025
• Mar 25: Gave a demo of MuJoCo Playground at GTC
• Feb 25: Open-sourced our latest project, MuJoCo Playground

Research

My research aims to make autonomous robots more capable and reliable in unstructured real-world environments. These days, I am focused on three key areas:

• Humanoid Control
  Making robots more useful by enabling them to learn a wide range of whole-body loco-manipulation skills.
• Data Priors
  Making robots look more human-like by leveraging prior data such as human demonstrations and motion capture.
• Simulation
  Leveraging large-scale simulation and careful system identification to create robust and generalizable controllers. Additionally, building an accessible ecosystem around simulation to foster science and reproducibility in robotics.

Publications

I make all my code available to support open and reproducible research. For a complete list, please see my Google Scholar.

• MuJoCo Playground - RSS 2025
  [Paper] [Website] [Code]
  Show BibTeX

  @misc{zakka2025mujocoplayground,
      title={MuJoCo Playground},
      author={Kevin Zakka and Baruch Tabanpour and Qiayuan Liao and Mustafa Haiderbhai and Samuel Holt and Jing Yuan Luo and Arthur Allshire and Erik Frey and Koushil Sreenath and Lueder A. Kahrs and Carmelo Sferrazza and Yuval Tassa and Pieter Abbeel},
      year={2025},
      eprint={2502.08844},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2502.08844},
  }

• ALOHA 2: An Enhanced Low-Cost Hardware for Bimanual Teleoperation - arXiv preprint, 2024
  [Paper] [Website] [Code]
  Show BibTeX

  @misc{aloha2,
      title={ALOHA 2: An Enhanced Low-Cost Hardware for Bimanual Teleoperation},
      author={ALOHA Team},
      year={2024},
      eprint={2405.02292},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2405.02292},
  }

• RoboPianist: Dexterous Piano Playing with Deep Reinforcement Learning - CoRL 2023
  [Paper] [Demo] [Code]
  Show BibTeX

  @inproceedings{robopianist2023,
      author = {Zakka, Kevin and Wu, Philipp and Smith, Laura and Gileadi, Nimrod and Howell, Taylor and Peng, Xue Bin and Singh, Sumeet and Tassa, Yuval and Florence, Pete and Zeng, Andy and Abbeel, Pieter},
      title = {RoboPianist: Dexterous Piano Playing with Deep Reinforcement Learning},
      booktitle = {Conference on Robot Learning (CoRL)},
      year = {2023},
  }

• XIRL: Cross-embodiment Inverse Reinforcement Learning - CoRL 2021
  ★ Best Paper Award Finalist
  [Paper] [Website] [Code]
  Show BibTeX

  @inproceedings{zakka2021xirl,
      title = {XIRL: Cross-embodiment Inverse Reinforcement Learning},
      author = {Zakka, Kevin and Zeng, Andy and Florence, Pete and Tompson, Jonathan and Bohg, Jeannette and Dwibedi, Debidatta},
      booktitle = {Conference on Robot Learning (CoRL)},
      year = {2021},
  }

• Form2Fit: Learning Shape Priors for Generalizable Assembly from Disassembly - ICRA 2020
  ★ Best Paper Award Finalist
  [Paper] [Code] [Website]
  Show BibTeX

  @inproceedings{zakka2020form2fit,
      title = {Form2Fit: Learning Shape Priors for Generalizable Assembly from Disassembly},
      author = {Zakka, Kevin and Zeng, Andy and Dwibedi, Debidatta and Bohg, Jeannette and Florence, Pete},
      booktitle = {International Conference on Robotics and Automation (ICRA)},
      year = {2020},
  }

Software

• mujoco_playground
  Train and deploy robot skills in minutes with GPU-accelerated MuJoCo — just pip install playground to get started
  
  
• mujoco_menagerie
  A growing collection of 50+ high-quality robot models for MuJoCo and beyond
  
  
• robot_descriptions.py
  Import and access 100+ robot descriptions in Python (contributed)
  
  
• mink
  Differential inverse kinematics in Python based on MuJoCo
  
  
• mjctrl
  Minimal, clean, single-file implementations of common robotics controllers in MuJoCo
  
  

New to MuJoCo or Robot Simulation?

I actively contribute to the MuJoCo community — building tools like mujoco_menagerie and mujoco_playground, and regularly helping out on GitHub issues and discussions.

If you're getting started with simulation or MuJoCo and have questions—whether it's debugging, guidance, or pointing you to the right resource—don't hesitate to reach out on GitHub or via email.

Teaching

• Graduate student instructor for CS 188: Introduction to AI in Fall 2024
• Head TA for CS 231n: Convolutional Neural Networks for Visual Recognition in Spring 2021

Media & Talks

Technical Writing

• Robot See, Robot Do (2022)
• CLIP: Zero-shot Jack of All Trades (2021)
• Representation Matters (2021)
• kNN classification using Neighbourhood Components Analysis (2020)
• Learning to Assemble and to Generalize from Self-Supervised Disassembly (2019)
• Deriving the gradient for the backward pass of batch normalization (2016)

Press Coverage

• MuJoCo Playground was mentioned in Jensen Huang's GTC keynote (March 2025)
  [Blog] [Video]

Awards & Service

• Centennial Teaching Assistant Award
  Stanford University, 2021 - [link]
• Conference Reviewing
  CoRL: 2023, 2024
  ICRA: 2020, 2022, 2023
  IROS: 2022, 2023, 2024

Industry Experience

• Boston Dynamics · Atlas Team Intern
  June - August 2024
  Supervised by Lucas Manuelli and Scott Kuindersma
• Google DeepMind · Simulation Team Intern
  June 2022 - May 2024
  Supervised by Yuval Tassa
• Google Robotics · Research Intern
  June - August 2019
  Supervised by Shuran Song and Andy Zeng

Contact

If you're interested in research discussions or collaboration, feel free to reach out:

• Email: zakka [at] berkeley [dot] edu