Jake Varley

Google (United States), Columbia University

Papers

11

Total Citations

142

H-Index

7

About

Jake Varley is a leading researcher at the intersection of large language models (LLMs) and robotic manipulation, whose work is fundamentally reshaping how robots plan, reason, and interact with the physical world. His primary research areas include uncertainty-aware planning, code-driven robot control, and scalable learning from offline data. Varley’s most impactful contribution is **KnowNo** (43 citations), a pioneering framework that enables robots to measure their own uncertainty and proactively ask for human help, directly addressing the critical problem of LLM hallucination in robotics. He also developed the influential **PromptBook for Code as Policies** (25 citations), which provides a systematic methodology for prompting LLMs to generate robot code, and **Actionable Models** (16 citations), a method for learning robust skills from pre-existing datasets without manual rewards. As a core contributor to **SARA-RT** and **Gemini Robotics**, Varley has advanced the scaling of transformer-based policies for real-world deployment. His work on bimanual manipulation and shared autonomy further demonstrates his commitment to creating safe, capable, and practical robotic systems that can operate alongside humans in complex environments.

Research Focus

Key Achievements

7
H-Index
11
Papers
142
Total Citations
13
Avg Citations/Paper
🏆 Most Cited Paper
Robots That Ask For Help: Uncertainty Alignment for Large Language Model Planners
43 citations · 2023
📈 Most Prolific Year: 2024 (4 Papers)
🤝 Key Collaborators: 143
🏛 Institutions: Google (United States), Columbia University

Top Papers

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10

Key Collaborators

Contact & Links

Available for collaboration
Content generated · 13 days ago