Rose Hendrix

Rose Hendrix is a robotics researcher whose work trains agents in simulation for navigation and manipulation in real settings. They coauthored a 2019 study of ergonomic risk in object handling and earned a Mechanical Engineering PhD at the University of Washington, where Santosh Devasia and Joseph Garbini advised them. Phone2Proc used a phone scan to build training scenes, while SPOC trained agents on simulated shortest paths for real-world navigation and manipulation. Molmo and PixMo introduced open-weight vision-language models and datasets; MolmoAct added action reasoning and won the 2025 People's Choice Award, while MolmoBot enabled zero-shot transfer to unseen objects and MolmoSpaces offered more than 230,000 indoor environments. As of October 2026, they are a research scientist on Ai2's Robotics team under Dieter Fox.

Now at: Allen Institute for AI (Ai2)
Works: 34
Citations: 3,524
h-index: 15

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

2019

Ergonomic risk research

In February 2019, they coauthor an arXiv study that predicts ergonomic risk from RGB-D videos of people manipulating objects. They earn a PhD at the University of Washington, advised by Santosh Devasia and Joseph Garbini.
2023

Phone2Proc at CVPR

In 2023, they coauthor Phone2Proc at CVPR with Matt Deitke, Kiana Ehsani and others. A 10-minute phone scan drives procedural training-scene generation, and ObjectNav success in sim-to-real evaluations rises from 34.7% to 70.7%.
2024

SPOC at CVPR

In 2024, they coauthor SPOC at CVPR, showing that agents imitating shortest-path planners in simulation can navigate and manipulate objects in the real world using only RGB sensors. Training uses millions of frames from about 200,000 procedurally generated houses.
September 2024

Molmo and PixMo

In September 2024, they coauthor Molmo and PixMo, an open-weight vision-language model family with open datasets. Its best 72B model outperforms larger proprietary models, including Claude 3.5 Sonnet and Gemini 1.5 Pro and Flash, second only to GPT-4o.
2025

MolmoAct action reasoning

In August 2025, they coauthor MolmoAct, an open action reasoning model that encodes observations and instructions into depth-aware perception tokens, spatial trajectory traces and low-level actions. It reaches 86.6% average success on LIBERO, and the project wins the People's Choice Award at the 2025 Allen School Research Showcase.
March 2026

MolmoBot zero-shot transfer

In March 2026, they coauthor MolmoBot, whose simulation-trained policies transfer zero-shot to unseen real-world objects and environments and reach 79.2% success on tabletop pick-and-place. The project wins Best Paper Awards at three ICRA 2026 workshops.
2026

MolmoSpaces indoor environments

In 2026, they coauthor MolmoSpaces, an open ecosystem for robot-policy benchmarking with more than 230,000 indoor environments.
As of October 2026

Research scientist at Ai2

They are a research scientist at the Allen Institute for AI on the Robotics team under Dieter Fox, working to scale robotics in simulation for broadly capable zero-shot generalist policies.