Rose Hendrix

rrceosnhob@tim.idrosxoec

I am currently working on robotics in stealth. Previously, I was a research scientist at the Allen Institute for Artificial Intelligence (Ai2) on the Robotics team under Dieter Fox, where I worked on massively scaling robotics in simulation to enable broadly capable zero-shot generalist policies. Before that, I earned my PhD in Mechanical Engineering at the University of Washington, where I was advised by Santosh Devasia and Joseph Garbini.

Google Scholar  /  Twitter

Research

I'm interested in how to bring robotics into unstructured human environments, particularly through the use of simulation and synthetic data. Here is my relevant recent work - please see my Google Scholar for a complete history. * denotes equal contribution.

MolmoB0T: Large-Scale Simulation Enables Zero-Shot Manipulation

MolmoB0T demonstrates that sufficiently large-scale and diverse simulated training data enables effective zero-shot sim-to-real transfer for both static and mobile manipulation, without any real-world fine-tuning. On tabletop pick-and-place, MolmoBot achieves 79.2% real-world success, outperforming π₀.₅ at 39.2%.
🏆 Best Paper Award at the Synthetic Data for Robot Learning (SDRL), VLA Pipeline, and Beyond Teleoperation workshops at ICRA 2026.

MolmoAct2: Action Reasoning Models for Real-world Deployment

MolmoAct2 is a fully open action reasoning model built for real-world deployment, with an embodied-reasoning VLM backbone (MolmoER), a flow-matching action expert, and adaptive-depth reasoning at a fraction of prior latency. It outperforms strong baselines including π₀.₅ across 7 simulation and real-world benchmarks, and releases the largest open bimanual dataset to date.
🏆 Best Paper Award at the TAG workshop at CoRL 2026.

MolmoSpaces: A Large-Scale Open Ecosystem for Robot Navigation and Manipulation

MolmoSpaces is a fully open ecosystem for large-scale robot policy benchmarking, featuring 230k+ diverse indoor environments, 130k annotated object assets with 42M stable grasps, and simulator-agnostic support for MuJoCo, Isaac, and ManiSkill. MolmoSpaces-Bench, its 8-task benchmark suite, exhibits strong sim-to-real correlation (R = 0.96, ρ = 0.98).

MolmoAct: Action Reasoning Models that can Reason in Space

MolmoAct encodes observations into depth-aware perception tokens, generates spatial plans as trajectory traces, and predicts precise actions, achieving 70.5% on SimplerEnv and 86.6% on LIBERO with explainable and steerable behavior.
🏆 Best Paper Runner-up at the Rational Robots workshop and Best Paper Nominee at the Data workshop at CoRL 2025.

GraspMolmo: Generalizable Task-Oriented Grasping via Large-Scale Synthetic Data Generation

GraspMolmo predicts semantically appropriate, stable grasps from natural language instructions and RGB-D frames. Trained on PRISM, a large-scale synthetic dataset of 379k samples, it achieves 70% success on complex real-world tasks compared to 35% for alternatives, with zero-shot bimanual grasping capabilities.

PoliFormer: On-Policy RL with Transformers Results in Masterful Navigators

Policy TransFormer (PoliFormer) is a transformer-based policy trained using RL at scale in simulation. PoliFormer achieves SoTA results across LoCoBot and Stretch RE-1, in both simulation and real-world.

Harmonic Mobile Manipulation

HarmonicMM is an end-to-end learning approach that combines navigation and manipulation, significantly improving success rates in complex tasks like door opening and table cleaning, with successful real-world transfer of agents trained in simulation.

Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Multimodal Models

Molmo and PixMo present open weights and open data to advance multimodal AI models, pushing the boundaries of state-of-the-art performance across various tasks.

FLaRe: Achieving Masterful and Adaptive Robot Policies with Large-Scale Reinforcement Learning Fine-Tuning

FLaRe utilizes large-scale reinforcement learning fine-tuning to create adaptive and highly capable robot policies, achieving state-of-the-art results in both simulated and real-world environments.

Open X-Embodiment: Robotic Learning Datasets and RT-X Models

We assemble a dataset from 22 different robots collected through a collaboration between 21 institutions, demonstrating 527 skills (160266 tasks).

SPOC: Imitating Shortest Paths in Simulation Enables Effective Navigation and Manipulation in the Real World

We train a supervised model to imitate shortest path trajectories collected from simulation and show that it generalizes to perform effective navigation and manipulation when deployed on real world agents.

Phone2Proc: Bringing Robust Robots Into Our Chaotic World

From a 10-minute iPhone scan of any environment, we condition procedural scene generation on that scan to generate training environments. Training a robot to perform ObjectNav in these scenes dramatically improves sim-to-real performance from 35% to 71% and results in an agent that is remarkably robust to human movement, lighting variations, added clutter, and rearranged objects.

Original design and source code from Jon Barron, modified by Leonid Keselman, and email scrambler by Jeff Donohue.