MolmoAct2: Action Reasoning Models for Real-world Deployment

Papers with Code Trending 05/04/26, 12:00 AM Models

robotics vla open-source action-reasoning allenai molmoact2

Summary

Allen AI releases MolmoAct2, an open-weight Vision-Language-Action model designed for real-world robotic deployment, featuring new datasets, an open action tokenizer, and adaptive reasoning to reduce latency.

Vision-Language-Action (VLA) models aim to provide a single generalist controller for robots, but today's systems fall short on the criteria that matter for real-world deployment. Frontier models are closed, open-weight alternatives are tied to expensive hardware, reasoning-augmented policies pay prohibitive latency for their grounding, and fine-tuned success rates remain below the threshold for dependable use. We present MolmoAct2, a fully open action reasoning model built for practical deployment, advancing its predecessor along five axes. We introduce MolmoER, a VLM backbone specialized for spatial and embodied reasoning, trained on a 3.3M-sample corpus with a specialize-then-rehearse recipe. We release three new datasets spanning low-to-medium cost platforms, including MolmoAct2-BimanualYAM, 720 hours of teleoperated bimanual trajectories that constitute the largest open bimanual dataset to date, together with quality-filtered Franka (DROID) and SO100/101 subsets. We provide OpenFAST, an open-weight, open-data action tokenizer trained on millions of trajectories across five embodiments. We redesign the architecture to graft a flow-matching continuous-action expert onto a discrete-token VLM via per-layer KV-cache conditioning. Finally, we propose MolmoThink, an adaptive-depth reasoning variant that re-predicts depth tokens only for scene regions that change between timesteps, retaining geometric grounding at a fraction of prior latency. In the most extensive empirical study of any open VLA to date, spanning 7 simulation and real-world benchmarks, MolmoAct2 outperforms strong baselines including Pi-05, while MolmoER surpasses GPT-5 and Gemini Robotics ER-1.5 across 13 embodied-reasoning benchmarks. We release model weights, training code, and complete training data. Project page: https://allenai.org/blog/molmoact2

Original Article

View Cached Full Text

Cached at: 05/08/26, 08:52 AM

Paper page - MolmoAct2: Action Reasoning Models for Real-world Deployment

Source: https://huggingface.co/papers/2605.02881 Published on May 4

Submitted byhttps://huggingface.co/Jiafei1224

Duanon May 5

#1 Paper of the day Authors:

Abstract

MolmoAct2 presents an open-action reasoning model for robotics that improves upon previous systems through specialized vision-language-model backbones, new datasets, open-weight action tokenizers, architectural redesign for continuous-action prediction, and adaptive reasoning for reduced latency.

Vision-Language-Action (VLA) models aim to provide a single generalist controller for robots, but today’s systems fall short on the criteria that matter for real-world deployment. Frontier models are closed, open-weight alternatives are tied to expensive hardware, reasoning-augmented policies pay prohibitive latency for their grounding, and fine-tuned success rates remain below the threshold for dependable use. We present MolmoAct2, a fully open action reasoning model built for practical deployment, advancing its predecessor along five axes. We introduce MolmoER, aVLM backbonespecialized for spatial andembodied reasoning, trained on a 3.3M-sample corpus with a specialize-then-rehearse recipe. We release three new datasets spanning low-to-medium cost platforms, including MolmoAct2-BimanualYAM, 720 hours of teleoperated bimanual trajectories that constitute the largest open bimanual dataset to date, together with quality-filtered Franka (DROID) and SO100/101 subsets. We provide OpenFAST, an open-weight, open-dataaction tokenizertrained on millions of trajectories across five embodiments. We redesign the architecture to graft aflow-matching continuous-action expertonto adiscrete-token VLMvia per-layerKV-cache conditioning. Finally, we propose MolmoThink, anadaptive-depth reasoningvariant that re-predicts depth tokens only for scene regions that change between timesteps, retaininggeometric groundingat a fraction of prior latency. In the most extensive empirical study of any open VLA to date, spanning 7 simulation and real-world benchmarks, MolmoAct2 outperforms strong baselines including Pi-05, while MolmoER surpasses GPT-5 and Gemini Robotics ER-1.5 across 13 embodied-reasoning benchmarks. We release model weights, training code, and complete training data. Project page: https://allenai.org/blog/molmoact2

View arXiv page View PDF Project page GitHub128 Add to collection

Get this paper in your agent:

hf papers read 2605\.02881

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash