Vision-language-action models are highly vulnerable to membership inference attacks, including practical black-box versions that exploit generated actions and motion trajectories.
super hub Canonical reference
RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control
Canonical reference. 77% of citing Pith papers cite this work as background.
256
Pith papers citing it
Background
77% of classified citations
abstract
We study how vision-language models trained on Internet-scale data can be incorporated directly into end-to-end robotic control to boost generalization and enable emergent semantic reasoning. Our goal is to enable a single end-to-end trained model to both learn to map robot observations to actions and enjoy the benefits of large-scale pretraining on language and vision-language data from the web. To this end, we propose to co-fine-tune state-of-the-art vision-language models on both robotic trajectory data and Internet-scale vision-language tasks, such as visual question answering. In contrast to other approaches, we propose a simple, general recipe to achieve this goal: in order to fit both natural language responses and robotic actions into the same format, we express the actions as text tokens and incorporate them directly into the training set of the model in the same way as natural language tokens. We refer to such category of models as vision-language-action models (VLA) and instantiate an example of such a model, which we call RT-2. Our extensive evaluation (6k evaluation trials) shows that our approach leads to performant robotic policies and enables RT-2 to obtain a range of emergent capabilities from Internet-scale training. This includes significantly improved generalization to novel objects, the ability to interpret commands not present in the robot training data (such as placing an object onto a particular number or icon), and the ability to perform rudimentary reasoning in response to user commands (such as picking up the smallest or largest object, or the one closest to another object). We further show that incorporating chain of thought reasoning allows RT-2 to perform multi-stage semantic reasoning, for example figuring out which object to pick up for use as an improvised hammer (a rock), or which type of drink is best suited for someone who is tired (an energy drink).
hub tools
citation-role summary
background 48
method 6
baseline 4
other 2
extension 1
citation-polarity summary
claims ledger
- abstract We study how vision-language models trained on Internet-scale data can be incorporated directly into end-to-end robotic control to boost generalization and enable emergent semantic reasoning. Our goal is to enable a single end-to-end trained model to both learn to map robot observations to actions and enjoy the benefits of large-scale pretraining on language and vision-language data from the web. To this end, we propose to co-fine-tune state-of-the-art vision-language models on both robotic trajectory data and Internet-scale vision-language tasks, such as visual question answering. In contrast
authors
co-cited works
representative citing papers
OSWorld provides the first unified real-computer benchmark for open-ended multimodal agent tasks, exposing large performance gaps between humans and state-of-the-art LLM/VLM agents.
Embodied.cpp introduces a portable C++ inference runtime with modular layers for deploying VLA and WAM models on heterogeneous robots, reporting 100% and 91% task success on two models plus memory reduction on a WAM benchmark.
Labimus is the first benchmark for humanoid dexterous manipulation in organic chemistry laboratories, exposing a gap between task completion and required experimental precision.
LIBERO-Safety supplies a scalable benchmark, data-generation pipeline, and 19,664-demonstration dataset that exposes a generalization-safety tension in current VLA models where diverse training improves collision avoidance but task success stays limited by trajectory quality and semantic understandi
RoboGaze presents a structured multi-agent VLM pipeline and robotics-specific error taxonomy that improves video evaluation metrics by up to 43 F1 points over zero-shot baselines on a 382-clip dataset.
A single Encoder-Router network uses semigroup superposition of frame, modulation, and coefficient parameters to produce a scene-specific Riemannian metric field that supports zero-shot geodesic planning after training on one two-obstacle scene.
UMI-Bench 1.0 is presented as the first open benchmark dedicated to reproducible real-world evaluation of Universal Manipulation Interface policies.
World models introduce a stealthy poisoning vector into robot learning pipelines where malicious prompts or dynamics in teleoperated data activate only during synthetic trajectory generation, enabling backdoors in downstream policies.
ActProbe is an action-space detector that uses temporal consistency error and action chunk magnitude from policy outputs, mapped via LSTM-MLP, to predict failures earlier than baselines across policies and real-robot tasks.
WIZARD meta-learns to map task evidence directly to LoRA updates for VLA policies, reporting up to 14x gains on unseen tasks in simulation and real-robot experiments without test-time optimization or action labels.
NextMotionQA benchmark reveals VLMs have critical gaps in fine-grained human motion understanding and align with experts on coarse judgment (κ=0.70) but not fine-grained (κ=0.10).
The paper identifies a deployment safety gap in VLA policies where identical checkpoints can be executable-inequivalent due to action metadata mismatches, supported by a derived closed-form transform and empirical drift measurements on LIBERO benchmarks.
TTT-VLA performs test-time training for VLA models by optimizing only a latent prompt on new interaction data via a proxy self-supervised signal, yielding higher task success rates on SimplerEnv in single- and multi-embodiment settings.
BOKBO is the first conformal abstention method for K-sample VLA policies that supplies finite-sample distribution-free guarantees on executed violation rates, with global and Mondrian per-task variants.
PhAIL provides an open benchmark and distributional evaluation method for real-robot VLA policies using time-to-success CDF, HRT scoring, and KS significance tests.
VLA architectures exhibit architecture-specific failure signatures at the motor-command level, with direction reversal as a universal predictor and velocity monitoring ineffective for continuous models.
Introduces Colosseum V2 benchmark for evaluating VLA model generalization in robotic manipulation with 28 tasks, revealing limitations in current methods and sim-real correlations.
Introduces MM-CreativityBench for affordance-grounded creative tool use and shows that DPO-based alignment with an affordance knowledge base improves entity and part selection while cutting hallucination errors in LMMs.
Visual CoT agents exhibit tool-use collapse where tool usage declines but task accuracy rises, and adding entropy regularization for rollout diversity produces the strongest performance.
CrossVLA develops a surrogate log-probability estimator for DPO on flow-matching VLAs, shows DoRA outperforming LoRA by +10.4 pp mean on LIBERO, and identifies inference bottlenecks with limited caching gains.
Demo-JEPA enables one-shot cross-embodiment imitation by mapping visual demonstrations to shared latent future trajectories that serve as subgoals for the target agent's own forward dynamics planning.
DSSP is a history-conditioned diffusion state space policy that uses SSMs to encode full observation streams with an auxiliary dynamics objective and hierarchical fusion, achieving SOTA results with reduced model size in robot manipulation.
A new VLA model called SI uses a four-step chain-of-thought to derive driving intent and applies it via classifier-free guidance to a flow-matching trajectory generator, showing competitive Waymo scores and intent-controllable plans.
citing papers explorer
-
OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments
OSWorld provides the first unified real-computer benchmark for open-ended multimodal agent tasks, exposing large performance gaps between humans and state-of-the-art LLM/VLM agents.
-
Advancing Creative Physical Intelligence in Large Multimodal Models
Introduces MM-CreativityBench for affordance-grounded creative tool use and shows that DPO-based alignment with an affordance knowledge base improves entity and part selection while cutting hallucination errors in LMMs.
-
Learning Interactive Real-World Simulators
UniSim learns a universal real-world simulator from orchestrated diverse datasets, enabling zero-shot deployment of policies trained purely in simulation.
-
Agentic RAG-VLM: Affordance-Aware Retrieval-Augmented Generation with Self-Reflective Planning for Robotic Grasping
Agentic RAG-VLM achieves 78.3% success on a 12-task grasping benchmark with 360 trials per configuration, a 53.3 percentage-point gain over VLM-only baselines, via hierarchical affordance RAG, scene graph constraints, and a 14-type failure taxonomy with adaptive retry.
-
DiBS: Diffusion-Informed Branch Selection
DiBS ranks candidate values with a diffusion model inside a symbolic Sudoku solver to reduce nodes, backtracks, and long-tail search cost on the Royle 17-clue benchmark.
-
Cosmos Policy: Fine-Tuning Video Models for Visuomotor Control and Planning
Single-stage fine-tuning of a video model to generate actions as latent frames plus future states and values yields state-of-the-art robot policy performance on LIBERO, RoboCasa, and bimanual tasks.
-
Block-wise Adaptive Caching for Accelerating Diffusion Policy
BAC accelerates transformer-based Diffusion Policy up to 3x by block-level adaptive feature caching using an Adaptive Caching Scheduler and Bubbling Union Algorithm to control error propagation.
-
V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning
V-JEPA 2 pre-trained on massive unlabeled video achieves strong results on motion understanding and action anticipation, SOTA video QA at 8B scale, and enables zero-shot robotic planning on Franka arms using only 62 hours of unlabeled robot video.
-
Cosmos-Reason1: From Physical Common Sense To Embodied Reasoning
Cosmos-Reason1-7B and 56B models are trained with physical common sense and embodied reasoning ontologies via supervised fine-tuning and reinforcement learning to produce next-step physical actions.
-
Preference Goal Tuning: Post-Training as Latent Control for Frozen Policies
PGT optimizes latent goal embeddings for frozen policies via trajectory-level preference objectives, reporting 72-81.6% relative gains on 17 Minecraft tasks and 13.4% better OOD performance than fine-tuning.
-
Cognitive Architectures for Language Agents
CoALA is a modular cognitive architecture for language agents that organizes memory components, action spaces for internal and external interaction, and a generalized decision-making loop to support more systematic development of capable agents.
-
Self-Evolving Cognitive Framework via Causal World Modeling for Embodied Scientific Intelligence
Proposes a self-evolving cognitive framework integrating causal world modeling, intervention-driven reasoning, and continual refinement for embodied scientific intelligence.
-
Anything2Skill: Compiling External Knowledge into Reusable Skills for Agents
Anything2Skill extracts and structures reusable skills from heterogeneous knowledge into a SkillBank that agents retrieve alongside RAG evidence, reaching 98.85% and 94.10% success on qsv and GitHub-CLI tasks.
-
Physically Native World Models: A Hamiltonian Perspective on Generative World Modeling
The paper introduces Hamiltonian World Models by encoding observations into structured latent phase space and evolving states via Hamiltonian-inspired dynamics for physically meaningful rollouts in embodied AI.
-
The Cartesian Cut in Agentic AI
LLM agents use a Cartesian split between learned prediction and engineered control, enabling modularity but creating sensitivity and bottlenecks unlike integrated biological systems.
-
LaGO: Latent Action Guidance for Online Reinforcement Learning
LaGO improves online RL success rates over vanilla PPO by using pretrained LLMs as latent action priors, raising rates from 15.1% to 27.2% on CLEVR-Robot and 2.7% to 15.2% on Meta-World.
-
Toward AI That Understands Self and Others: A World-Model Theory of Cognitive Diversity and Alignment
The paper introduces the Multi-Phase Inference Assumption and Mechanism to frame cognitive diversity as arising from constrained construction of sufficient statistics and defines alignment as processability between heterogeneous world models via alignment maps and transformation loss.
-
Aligning Perception, Reasoning, Modeling and Interaction: A Survey on Physical AI
A survey of physical AI that distinguishes theoretical physics reasoning from applied understanding and synthesizes advances in symbolic reasoning, embodied systems, and generative models to advocate for physics-grounded world models.
-
Agent AI: Surveying the Horizons of Multimodal Interaction
The paper defines Agent AI as interactive multimodal systems that perceive grounded data and generate embodied actions, arguing this approach can mitigate hallucinations in foundation models.