Rho models improve robot hand coordination and task adaptation
Rho: A Foundation for Efficiently Adaptable VLA Models
Robotics
Summary
Making robots with two arms work well on many tasks is hard because each robot looks and moves differently. The authors created Rho, a set of robot control models that work with several popular two-arm robots. These models learn well even with little new data and can quickly adjust to new tasks. They tested Rho both in simulations and real robots, showing it matches or beats existing systems. Rho can also adapt on the fly by learning from human corrections, which helps it handle tricky situations it didn’t see before.
What this means in practice
- •For robotics engineers: Use Rho’s adaptable models to efficiently program dual-arm robots for new manipulation tasks across different robot types with minimal new data.
- •For industrial automation teams: Deploy Rho’s pre-trained robot control checkpoints to speed up integration and fine-tuning of bimanual robots in manufacturing or warehouse settings.$Commercial implications: Rho enables broader dual-arm robot deployment by reducing adaptation time and labor costs, making customizable robot workcells commercially viable.
Authors
Rho Team, Simran Bagaria, Daphne Chen, Dean Fortier, Jianlong Fu, Michael Harrison, Tess Hellebrekers, Neel Joshi, Andrey Kolobov, Dalton Moore, Galen Mullins, Michael Murray, Eduardo Salinas, Reuben Tan
Abstract
General-purpose physical AI models must combine broad visual and linguistic capabilities with precise control across robot embodiments and efficient adaptation to downstream tasks. We introduce Rho, a family of open-weights VLA models for bimanual manipulation designed for data-light task adaptation on 3 embodiments representative of dual-arm robots across research labs and the industry -- YAM Box, UR AI Trainer, and FR3 Duo. We systematically ablate Rho's action-expert architecture and training recipe, and show in controlled simulation and physical-robot experiments that embodiment midtraining improves downstream adaptation. The resulting Rho variants for YAM Box, UR AI Trainer, and FR3 Duo match or outperform existing open-weights VLAs and achieve the strongest overall performance across the tasks, embodiments, and baselines evaluated in this report. We further demonstrate the Rho model family's built-in capacity for online adaptation: an internal latent policy learns from corrective feedback to select observation-conditioned noise inputs for the frozen flow-matching action expert. With as few as 15 corrected episodes, adapting this lightweight module enables Rho to handle task situations at the fringe of its offline finetuning distribution. Together, these results position Rho as both a strong general-purpose robotic manipulation model and a practical foundation for adaptation. We release the base Rho model and the embodiment-specific checkpoints to facilitate Rho's deployment in research experiments and practical industrial use cases.