# Vision-Language-Action model

7 related articles

Subscribe to updates

arXiv Nov 7, 2026
Action generation stops stitching chunks: ODEAction evolves actions by ODE and lifts EBench average success rate by 61.8% relative
Vision-language-action models usually predict a whole future action chunk in one pass, and adjacent chunks lack explicit temporal consistency constraints, so actions can jump at chunk boundaries. ODEAction recasts generation as continuous evolution from the current action: it learns an action transition field and integrates it with an ordinary differential equation to produce subsequent actions, instead of jointly predicting a fixed-length chunk. Average success rate rises from 96.85% to 98.25% on LIBERO and from 15.28% to 24.72% on EBench, a 61.8% relative gain; on five RoboTwin tasks it reaches 84.4% average success against the 82.4% baseline.
Vision-language-action models usually predict a whole future action chunk in one pass, and adjacent chunks lack explicit temporal consistency constraints, so actions can jump at chunk boundaries. ODEAction recasts generation as continuous evolution from the current action: it learns an action transition field and integrates it with an ordinary differential equation to produce subsequent actions, instead of jointly predicting a fixed-length chunk. Average success rate rises from 96.85% to 98.25% on LIBERO and from 15.28% to 24.72% on EBench, a 61.8% relative gain; on five RoboTwin tasks it reaches 84.4% average success against the 82.4% baseline.
Central University of Finance and Economics Fudan University Tsinghua University Shenzhen University of Advanced Technology Renmin University of China Beijing University of Posts and Telecommunications Westlake University
Qishen Wang（First author）
ODEAction Continuous Action Evolution Vision-Language-Action model Action Chunking
arXiv Oct 15, 2026
VLA action chunks get a validity check before execution: PAV predicts each prefix's outcome and lifts LIBERO average success to 94.65%
Vision-language-action (VLA) policies emit a whole action sequence per query, but how much of it to execute was previously judged only indirectly, through generation stability or prediction agreement. PAV instead predicts, inside the policy's own task-conditioned representation, what a candidate prefix will lead to, and a verifier trained on trajectory-level outcomes scores whether the task can still be completed. On the four LIBERO suites, PAV averages 94.65% success at 14.0 policy queries per episode, above direct verification without future prediction (84.55%, 22.3 queries) and prediction in DINOv2 features (87.00%, 22.2 queries).
Vision-language-action (VLA) policies emit a whole action sequence per query, but how much of it to execute was previously judged only indirectly, through generation stability or prediction agreement. PAV instead predicts, inside the policy's own task-conditioned representation, what a candidate prefix will lead to, and a verifier trained on trajectory-level outcomes scores whether the task can still be completed. On the four LIBERO suites, PAV averages 94.65% success at 14.0 policy queries per episode, above direct verification without future prediction (84.55%, 22.3 queries) and prediction in DINOv2 features (87.00%, 22.2 queries).
Seoul National University Korea Advanced Institute of Science and Technology
Minkyu Kim（First author）
Policy-Aligned Predictive Verifier Adaptive Action-Chunk Execution Vision-Language-Action model Action Chunking LIBERO
arXiv Oct 8, 2026
DRIVE turns successful-trajectory diversity into an RL objective, improving VLA out-of-domain generalization in simulation and on real robots
The authors analyze how reinforcement learning fine-tuning (RFT) reshapes exploration in vision-language-action (VLA) policies, finding that RFT contracts the overall behavior distribution while making successful trajectories more diverse, easier to elicit, and broader in task-valid solution coverage; they then propose DRIVE, which converts temporally aligned similarity among grouped rollouts under matched task conditions into a success-conditioned intrinsic reward, improving out-of-domain success on LIBERO-Plus, ManiSkill3, and RoboTwin 2.0 and raising average out-of-domain success from 64.1% to 73.3% on a dual-arm AgileX PiPER-X platform.
The authors analyze how reinforcement learning fine-tuning (RFT) reshapes exploration in vision-language-action (VLA) policies, finding that RFT contracts the overall behavior distribution while making successful trajectories more diverse, easier to elicit, and broader in task-valid solution coverage; they then propose DRIVE, which converts temporally aligned similarity among grouped rollouts under matched task conditions into a success-conditioned intrinsic reward, improving out-of-domain success on LIBERO-Plus, ManiSkill3, and RoboTwin 2.0 and raising average out-of-domain success from 64.1% to 73.3% on a dual-arm AgileX PiPER-X platform.
Australian National University Nanjing University Harbin Institute of Technology University of Technology Sydney
Haoru Li（First author）
Diversity-Driven RL Fine-Tuning Vision-Language-Action model Global Alignment Kernel Potential-Based Reward Shaping Proximal Policy Optimization
arXiv Oct 1, 2026
RL post-training of vision-language-action models concentrates parameter updates in Timestep Modules that hold only 14.83%–27.58% of the action expert, and their low-rank directions predict and improve task success
This work systematically analyzes how reinforcement learning (RL) reshapes flow-based vision-language-action (VLA) models, finding that across π0.5 and GR00T N1.5/N1.6 on LIBERO, ManiSkill, MetaWorld, and CALVIN, RL induces low-rank parameter updates highly concentrated in the action expert's Timestep Modules, which hold 14.83%–27.58% of the parameters yet capture a disproportionate share of the RL gain, with shift-vector update directions predicting task success at up to 99.6% ROC-AUC and steering along them improving policies without additional RL training.
This work systematically analyzes how reinforcement learning (RL) reshapes flow-based vision-language-action (VLA) models, finding that across π0.5 and GR00T N1.5/N1.6 on LIBERO, ManiSkill, MetaWorld, and CALVIN, RL induces low-rank parameter updates highly concentrated in the action expert's Timestep Modules, which hold 14.83%–27.58% of the parameters yet capture a disproportionate share of the RL gain, with shift-vector update directions predicting task success at up to 99.6% ROC-AUC and steering along them improving policies without additional RL training.
Minjae Oh（First author）
Vision-Language-Action model RL Post-Training Low-Rank Parameter Updates Timestep Modules Activation Steering
arXiv Sep 29, 2026
A strong agent writes its intervention experience into a playbook for a light agent, lifting real-world manipulation success from 37.3% to 64.0%
The work proposes Recursive Harness Distillation (RHD): a strong agent (GPT-6 Astra) first interacts with a frozen vision-language-action model and distills effective intervention strategies into a reusable playbook, then recursively revises that playbook using the light agent's (GPT-5.6 Luna) execution feedback, improving manipulation without updating any model parameters—real-world success rises from 37.3% to 64.0%, the light agent with the playbook reaches 66.7% on SimplerEnv Bridge versus 41.7% for GR00T alone, and the same playbook also lifts the strong agent to 79.2%.
The work proposes Recursive Harness Distillation (RHD): a strong agent (GPT-6 Astra) first interacts with a frozen vision-language-action model and distills effective intervention strategies into a reusable playbook, then recursively revises that playbook using the light agent's (GPT-5.6 Luna) execution feedback, improving manipulation without updating any model parameters—real-world success rises from 37.3% to 64.0%, the light agent with the playbook reaches 66.7% on SimplerEnv Bridge versus 41.7% for GR00T alone, and the same playbook also lifts the strong agent to 79.2%.
Seoul National University Korea Advanced Institute of Science and Technology
Seungyeon Kim（First author）
Recursive Harness Distillation Vision-Language-Action model Robot Manipulation
arXiv Sep 28, 2026
HuRo turns five human-video sources into about 142 million robotized frames, lifting real-robot OOD completion from 34.9% to 72.2% in ALLEX VLA pretraining
HuRo introduces a robotization pipeline that converts heterogeneous egocentric human videos into robot-aligned observations and action trajectories, builds the HuRo dataset of about 142 million frames from five sources (EgoDex, EgoVerse, Ego4D, Ego10K, and EPIC-Kitchens), and uses it to pretrain VLA policies on the ALLEX bimanual dexterous robot: as robotized data increases, average completion across four real-world manipulation tasks rises from 51.5% without pretraining to 80.3% with the full dataset, with ID completion from 68.1% to 88.4% and OOD completion from 34.9% to 72.2%.
HuRo introduces a robotization pipeline that converts heterogeneous egocentric human videos into robot-aligned observations and action trajectories, builds the HuRo dataset of about 142 million frames from five sources (EgoDex, EgoVerse, Ego4D, Ego10K, and EPIC-Kitchens), and uses it to pretrain VLA policies on the ALLEX bimanual dexterous robot: as robotized data increases, average completion across four real-world manipulation tasks rises from 51.5% without pretraining to 80.3% with the full dataset, with ID completion from 68.1% to 88.4% and OOD completion from 34.9% to 72.2%.
Jinho Jeong（First author）
HuRo Human-video robotization pipeline VLA Pretraining Vision-Language-Action model Motion Retargeting
arXiv Sep 26, 2026
ART turns VLA models into tool-calling robot agents, reporting a 20% higher success rate than mainstream baselines in simulation and real-world tasks
The work proposes Agentic Robot with Tool-use (ART), a tool-injection framework that tunes any VLA model to leverage off-the-shelf tool modules for low-level vision, high-level affordance, and embodiment enhancement; the authors built a dataset of 30K tool-use trajectories and action demonstrations and designed a training regimen for long-trajectory tool-use reasoning in challenging environments, with experiments showing ART achieves a 20% higher success rate than mainstream baselines on simulation and real-world tasks such as pick-and-place in the dark at novel viewpoints.
The work proposes Agentic Robot with Tool-use (ART), a tool-injection framework that tunes any VLA model to leverage off-the-shelf tool modules for low-level vision, high-level affordance, and embodiment enhancement; the authors built a dataset of 30K tool-use trajectories and action demonstrations and designed a training regimen for long-trajectory tool-use reasoning in challenging environments, with experiments showing ART achieves a 20% higher success rate than mainstream baselines on simulation and real-world tasks such as pick-and-place in the dark at novel viewpoints.
Chinese University of Hong Kong Tsinghua University
Yi Ding（First author）
ART (Agentic Robot with Tool-use)agentic tool-use Vision-Language-Action model Robot Manipulation

You’re all caught up

Action generation stops stitching chunks: ODEAction evolves actions by ODE and lifts EBench average success rate by 61.8% relative

VLA action chunks get a validity check before execution: PAV predicts each prefix's outcome and lifts LIBERO average success to 94.65%

DRIVE turns successful-trajectory diversity into an RL objective, improving VLA out-of-domain generalization in simulation and on real robots

RL post-training of vision-language-action models concentrates parameter updates in Timestep Modules that hold only 14.83%–27.58% of the action expert, and their low-rank directions predict and improve task success

A strong agent writes its intervention experience into a playbook for a light agent, lifting real-world manipulation success from 37.3% to 64.0%

HuRo turns five human-video sources into about 142 million robotized frames, lifting real-robot OOD completion from 34.9% to 72.2% in ALLEX VLA pretraining

ART turns VLA models into tool-calling robot agents, reporting a 20% higher success rate than mainstream baselines in simulation and real-world tasks