Jake Varley
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
Top Papers
- 1
- 2
- 3
- 4
- 5Task level hierarchical system for BCI-enabled shared autonomy14 citations · 2017
- 6Embodied AI with Two Arms: Zero-shot Learning, Safety and Modularity10 citations · 2024
- 7
- 8Scaling Up Multi-Task Robotic Reinforcement Learning4 citations · 2021
- 9Gemini Robotics: Bringing AI into the Physical World4 citations · 2025
- 10