Tags: Vision-Language-Action Model - Paper Library

REALM: A Real-to-Sim Validated Benchmark for Generalization in Robotic Manipulation

Published:12/23/2025

Vision-Language-Action ModelRobotic Manipulation BenchmarkRobotic Generalization EvaluationHigh-Fidelity Simulation EnvironmentTask Perturbation Factors

REALM introduces a highfidelity simulation environment for evaluating the generalization of VisionLanguageAction models, featuring 15 perturbation factors and over 3,500 objects. Validation shows strong alignment between simulated and realworld performance, highlighting ongoi

Actions as Language: Fine-Tuning VLMs into VLAs Without Catastrophic Forgetting

Published:9/26/2025

Fine-Tuning of Visual Language ModelsVision-Language-Action ModelCatastrophic Forgetting PreventionLow-Rank Adaptation MethodRobot Teleoperation Data

The paper presents VLM2VLA, a method for finetuning visionlanguage models (VLMs) into visionlanguageaction models (VLAs) without catastrophic forgetting, by representing lowlevel robot actions in natural language, achieving zeroshot generalization in real experiments.

ReBot: Scaling Robot Learning with Real-to-Sim-to-Real Robotic Video Synthesis

Published:3/16/2025

Vision-Language-Action ModelRobotic Video SynthesisReal-to-Sim-to-Real ApproachRobot Dataset ScalingRobotic Manipulation Tasks

ReBot enhances robot learning by proposing a realtosimtoreal video synthesis method, addressing data scaling challenges. It replays real robot movements in simulators and combines them with inpainted realworld backgrounds, significantly improving VLA model performance with s

MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation

Published:8/27/2025

Vision-Language-Action ModelRobotic ManipulationLong-Term Memory and Anticipatory ActionMemory-Conditioned Diffusion ModelsShort-Term Memory and Cognition Fusion

MemoryVLA is a memorycentric VisionLanguageAction framework for nonMarkovian robotic manipulation, integrating working memory and episodic memory. It significantly enhances performance in 150 tasks, achieving up to a 26% success rate increase across simulations and realworl

Real-Time Execution of Action Chunking Flow Policies

Published:6/9/2025

Real-Time Action Chunking Policy ExecutionVision-Language-Action ModelHigh-Frequency Control TasksKinetix SimulatorAction Chunking Algorithm

This paper introduces a novel algorithm called realtime chunking (RTC) to address inference latency issues in realtime control of visionlanguageaction models, showing improved task throughput and high success rates in dynamic and realworld tasks.

VLA-4D: Embedding 4D Awareness into Vision-Language-Action Models for SpatioTemporally Coherent Robotic Manipulation

Published:11/21/2025

Vision-Language-Action ModelSpatiotemporal Coherent Robotic Manipulation4D-aware Visual RepresentationMultimodal Action RepresentationVLA Dataset Extension

The VLA4D model introduces 4D awareness into VisionLanguageAction models for coherent robotic manipulation, integrating spatial and temporal information to ensure smooth and consistent actions in robot tasks.

See Once, Then Act: Vision-Language-Action Model with Task Learning from One-Shot Video Demonstrations

Published:12/8/2025

Vision-Language-Action ModelRobotic Manipulation Policy LearningOne-Shot Learning from DemonstrationLearning from Human Video DemonstrationsExpert Demonstration Video Generation

ViVLA is a general robotic manipulation model that learns new tasks from a single expert video demonstration. By processing the video alongside robot observations, it distills expertise for improved performance in unseen tasks, showing significant experimental gains.

GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

Published:3/19/2025

Generalist Humanoid Robot Foundation ModelVision-Language-Action ModelDiffusion Transformer ModuleHumanoid Robot Manipulation TasksMultimodal Data Training

GR00T N1 is an open foundation model for humanoid robots, integrating a reasoning module and a motion generation module. Trained endtoend with a pyramid of heterogeneous data, it outperforms existing imitation learning methods in simulation benchmarks, demonstrating high perfor

FailSafe: Reasoning and Recovery from Failures in Vision-Language-Action Models

Published:10/2/2025

Vision-Language-Action ModelRobotic Manipulation Failure RecoveryFailure Generation and Recovery SystemRobotic Manipulation DatasetLarge-Scale Robot Training Data

The paper introduces FailSafe, a system that generates diverse failure scenarios and executable recovery actions for VisionLanguageAction (VLA) models, achieving up to 22.6% performance improvement in robotic failure recovery across various tasks.

RoboFAC: A Comprehensive Framework for Robotic Failure Analysis and Correction

Published:5/18/2025

Robotic Failure Analysis and Correction FrameworkVision-Language-Action ModelTask Understanding for Failure CorrectionRoboFAC DatasetOpen-World Robotic Manipulation

The RoboFAC framework enhances robotic failure analysis and correction for VisionLanguageAction models in openworld scenarios. It includes a large dataset and a model capable of task understanding, with experimental results showing significant performance improvements.

TrackVLA++: Unleashing Reasoning and Memory Capabilities in VLA Models for Embodied Visual Tracking

Published:10/8/2025

Vision-Language-Action ModelSpatial Reasoning MechanismTarget Identification MemoryLong-Horizon Consistency ModelingAutoregressive Reasoning Model

TrackVLA is a novel VisionLanguageAction model that enhances embodied visual tracking by introducing a spatial reasoning mechanism and Target Identification Memory. It effectively addresses tracking failures under severe occlusions and achieves stateoftheart performance.

011

SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning

Published:9/12/2025

Vision-Language-Action ModelReinforcement Learning for Math ReasoningRL Training for Large Language ModelsMulti-Environment RenderingEfficient Reinforcement Learning Framework

SimpleVLARL is introduced to enhance the training of VisionLanguageAction models using reinforcement learning, addressing data scarcity and generalization issues. Results show stateoftheart performance on OpenVLAOFT, reducing reliance on labeled data.

Spatial Forcing: Implicit Spatial Representation Alignment for Vision-language-action Model

Published:10/14/2025

Vision-Language-Action ModelEnhanced Spatial Understanding CapabilitiesImplicit Spatial Representation AlignmentAlignment with 3D Foundation ModelsPrecise Execution of Robotic Tasks

This paper introduces 'Spatial Forcing' (SF), an implicit alignment method enhancing spatial understanding in VisionLanguageAction (VLA) models. By aligning visual embeddings with 3D foundation models, SF improves robotics' operational precision in 3D environments without relyi

X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model

Published:10/12/2025

Vision-Language-Action ModelCross-Embodiment LearningSoft Prompt LearningGeneralist Robotic PlatformsLarge-Scale Heterogeneous Datasets

The paper introduces XVLA, a scalable VisionLanguageAction model utilizing a softprompted Transformer architecture. By integrating learnable embeddings for diverse robot data sources, XVLA achieves stateoftheart performance across simulations and real robots, demonstratin

FAST: Efficient Action Tokenization for Vision-Language-Action Models

Published:1/17/2025

Vision-Language-Action ModelFrequency-space Action Sequence TokenizationHigh-Frequency Robot Action DataAutoregressive Sequence ModelsRobot Action Tokenization

This paper introduces Frequencyspace Action Sequence Tokenization (FAST), enhancing autoregressive visionlanguageaction models for highfrequency robot tasks. FAST allows effective handling of challenging dexterous actions, releasing FAST, a versatile tokenizer that reduces t

ADriver-I: A General World Model for Autonomous Driving

Published:11/23/2023

Driving World ModelsMultimodal Large Language ModelVision-Language-Action ModelDiffusion ModelsnuScenes Dataset

ADriverI integrates visionaction pairs using MLLM and diffusion models to autoregressively predict control signals and future scenes, enabling iterative autonomous driving and significantly improving performance.

TrackVLA: Embodied Visual Tracking in the Wild

Published:5/29/2025

Vision-Language-Action ModelEmbodied Visual TrackingDiffusion Model for Trajectory PlanningLLM BackboneEmbodied Visual Tracking Benchmark (EVT-Bench)

TrackVLA integrates vision, language, and action via a shared LLM, jointly optimizing target recognition and trajectory planning with diffusion models. Trained on a large EVTBench dataset, it achieves stateoftheart, robust embodied visual tracking in complex realworld scenar

Emergent Active Perception and Dexterity of Simulated Humanoids from Visual Reinforcement Learning

Published:5/18/2025

Vision-Language-Action ModelRobotic Action LearningLLM-guided motion planningReinforcement Learning TrainingSimulated Humanoid Control

This work introduces PDC, enabling simulated humanoids to perform multiple tasks using egocentric vision alone. Reinforcement learning yields emergent humanlike behaviors such as active search, advancing visiondriven dexterous control without privileged states.

010

PhysVLM: Enabling Visual Language Models to Understand Robotic Physical Reachability

Published:3/11/2025

Vision-Language-Action ModelRobotic Physical Reachability RepresentationMulti-Robot Multimodal Dataset Phys100KRobotic Vision-Language ReasoningEnvironmental Perception and Spatial Representation

PhysVLM integrates a unified SpacePhysical Reachability Map into visionlanguage models, enabling accurate physical reachability reasoning for robots. It enhances embodied visual reasoning without compromising visionlanguage capabilities, validated on the largescale multirobo

WebWatcher: Breaking New Frontiers of Vision-Language Deep Research Agent

Published:10/8/2025

Multimodal Large Language ModelVision-Language-Action ModelRL Training for Large Language ModelsComplex Information Retrieval BenchmarkVisual-Language Reasoning

WebWatcher is a multimodal deep research agent enhancing visuallanguage reasoning via synthetic trajectories and reinforcement learning, validated on the new BrowseCompVL benchmark for complex visualtext retrieval tasks, surpassing existing baselines.

1 - 20 / 22

Papers