Kevin Zakka

podcast  /  twitter  /  blog  /  github  /  scholar  /  linkedin

I'm a PhD student at UC Berkeley, advised by Pieter Abbeel.

I recently graduated from Stanford University with a distinction in research.

During my masters, I was an intern in the Robotics team at Google Brain and an AI resident at Everyday Robots.

Research

My research centers around robotic perception and manipulation. In particular, I'm interested in endowing robots with the flexible ability to learn from rich sources of data, like video demonstrations and natural language instructions.

Publications

• XIRL: Cross-embodiment Inverse Reinforcement Learning
  Kevin Zakka, Andy Zeng, Pete Florence, Jonathan Tompson, Jeannette Bohg, Debdidatta Dwibedi
  Conference on Robot Learning (CoRL), 2021 (Oral Presentation)
  ★ Best Paper Award Finalist
  [arXiv] [code] [website]
• Form2Fit: Learning Shape Priors for Generalizable Assembly from Disassembly
  Kevin Zakka, Andy Zeng, Johnny Lee, Shuran Song
  IEEE International Conference on Robotics and Automation (ICRA) 2020
  ★ Best Paper Award in Automation Finalist
  [arXiv] [code] [website]

Software

I actively contribute to MuJoCo, a popular and widely used physics simulator for robotics and control.

Some representative software I've developed over the years:

• MuJoCo Menagerie: a collection of high-quality models for MuJoCo, curated by DeepMind
• obj2mjcf: a tool for converting OBJ files into multiple MuJoCo meshes grouped by material
• dexterity: software and tasks for dexterous hand manipulation, powered by MuJoCo
• CLIP Playground: colab notebooks showcasing CLIP's impressive zero-shot capabilities
• learn-linalg: numerical linear algebra routines from scratch in pure python

Teaching

• CS231n: Convolutional Neural Networks for Visual Recognition
  Head TA (Spring 2021) & TA (Spring 2020)
  Awarded Centennial Teaching Assistant for contributions to the class