Scholar
Karl Pertsch
Google Scholar ID: 3oe0I0QAAAAJ
UC Berkeley, Stanford University
Citations & Impact
Citations
8,176
H-index
25
i10-index
31
Publications
20
Co-authors
10
list available
Publications
Resume (English only)
Academic Achievements
- Published 'FAST: Efficient Action Tokenization for Vision-Language-Action Models' (ArXiv, 2025): introduced a simple, efficient action tokenization method enabling 5x faster VLA training and zero-shot generalization across environments.
- Co-authored 'Re-Mix: Optimizing Data Mixtures for Large Scale Imitation Learning' (CoRL 2024): developed a scalable data mixture optimization method using group distributionally robust optimization, outperforming expert-tuned weights on RT-X.
- Co-authored 'Robotic Control via Embodied Chain-of-Thought Reasoning' (CoRL 2024): proposed embodied chain-of-thought learning for VLAs, improving OpenVLA generalization by 30% without additional robot data.
- Co-first author of 'OpenVLA: An Open-Source Vision-Language-Action Model' (CoRL 2024): released a 7B-parameter open-source VLA pretrained on 970k robot episodes, achieving state-of-the-art generalist manipulation and supporting multi-robot control and efficient fine-tuning.
- Contributed to 'Evaluating Real-World Robot Manipulation Policies in Simulation' (CoRL 2024): introduced SIMPLER, a suite of simulated environments for policy evaluation.
