Recognition: 2 theorem links
RT-1: Robotics Transformer for Real-World Control at Scale
Pith reviewed 2026-05-10 22:36 UTC · model grok-4.3
The pith
Training a high-capacity transformer on large diverse real-robot datasets produces generalization to new tasks.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
We present a model class, dubbed Robotics Transformer, that exhibits promising scalable model properties. We verify our conclusions in a study of different model classes and their ability to generalize as a function of the data size, model size, and data diversity based on a large-scale data collection on real robots performing real-world tasks.
What carries the argument
The Robotics Transformer, a transformer architecture trained task-agnostically on real-robot interaction data that absorbs scale in data volume, model capacity, and task diversity to improve downstream control performance.
If this is right
- Performance on unseen tasks improves measurably when more real-robot demonstrations are added to training.
- Larger model capacity yields better generalization when the data mixture stays diverse.
- Task diversity during pretraining contributes to robustness beyond raw data volume alone.
- A single model trained this way can be deployed across multiple control problems without task-specific retraining.
Where Pith is reading between the lines
- Robot deployment pipelines could shift from collecting task-specific data to curating broad experience corpora that serve many applications.
- The same scaling pattern might let models trained on robot data plus video or simulation sources handle even wider real-world variation.
- Data collection efforts in robotics would benefit from prioritizing coverage across environments and objects rather than depth in narrow skills.
Load-bearing premise
That open-ended task-agnostic training on diverse real-robot data with a high-capacity model will continue to produce better generalization as scale increases.
What would settle it
A controlled experiment in which increasing dataset size, model size, or task diversity produces no further gains or causes worse performance on new real-robot tasks.
read the original abstract
By transferring knowledge from large, diverse, task-agnostic datasets, modern machine learning models can solve specific downstream tasks either zero-shot or with small task-specific datasets to a high level of performance. While this capability has been demonstrated in other fields such as computer vision, natural language processing or speech recognition, it remains to be shown in robotics, where the generalization capabilities of the models are particularly critical due to the difficulty of collecting real-world robotic data. We argue that one of the keys to the success of such general robotic models lies with open-ended task-agnostic training, combined with high-capacity architectures that can absorb all of the diverse, robotic data. In this paper, we present a model class, dubbed Robotics Transformer, that exhibits promising scalable model properties. We verify our conclusions in a study of different model classes and their ability to generalize as a function of the data size, model size, and data diversity based on a large-scale data collection on real robots performing real-world tasks. The project's website and videos can be found at robotics-transformer1.github.io
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript introduces the Robotics Transformer (RT-1), a high-capacity transformer-based model trained in an open-ended, task-agnostic manner on large-scale collections of real-robot trajectories. It claims that this combination of training regime and architecture yields promising scalable properties, with generalization improving as a function of data volume, model capacity, and data diversity; these trends are verified through controlled experiments on real robots executing diverse real-world tasks.
Significance. If the reported scaling trends hold under further scrutiny, the work would be significant for robotics by providing the first large-scale empirical demonstration that lessons from vision and language scaling can transfer to embodied control. The explicit study of data size, model size, and diversity on a real-robot corpus is a strength, as is the release of the project website with videos for qualitative inspection.
minor comments (3)
- Abstract: no quantitative success rates, error bars, or baseline comparisons are stated, forcing readers to reach the full results section before assessing the strength of the 'promising scalable properties' claim.
- The manuscript would benefit from an explicit statement of the total number of real-robot trajectories, the number of distinct tasks, and the precise train/test split protocol used for the scaling ablations.
- Figure captions and axis labels in the scaling plots should include the exact model sizes (parameter counts) and data volumes corresponding to each point to allow direct replication of the curves.
Simulated Author's Rebuttal
We thank the referee for the positive and accurate summary of our work, as well as for the recommendation of minor revision. We are pleased that the significance of demonstrating scalable generalization properties for embodied control via large-scale, task-agnostic training on real-robot data is recognized, along with the value of the controlled experiments on data volume, model capacity, and diversity.
Circularity Check
No significant circularity; purely empirical scaling study
full rationale
The paper presents the Robotics Transformer as a model class and verifies its scalable properties through direct empirical experiments on real-robot data, varying data size, model size, and diversity. No mathematical derivations, equations, predictions, or first-principles results exist that could reduce to inputs by construction. The argument for task-agnostic training plus high-capacity architectures is a hypothesis tested by the scaling studies rather than derived from self-citations or definitions. No self-definitional, fitted-input, uniqueness-imported, or ansatz-smuggled steps are present. The work is self-contained as a standard empirical scaling analysis.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption High-capacity architectures can absorb diverse robotic data to enable generalization
Forward citations
Cited by 60 Pith papers
-
Membership Inference Attacks on Vision-Language-Action Models
Vision-language-action models are highly vulnerable to membership inference attacks, including practical black-box versions that exploit generated actions and motion trajectories.
-
FlowHijack: A Dynamics-Aware Backdoor Attack on Flow-Matching Vision-Language-Action Models
FlowHijack is the first dynamics-aware backdoor attack on flow-matching VLAs that achieves high success rates with stealthy triggers while preserving benign performance and making malicious actions kinematically indis...
-
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.
-
RotVLA: Rotational Latent Action for Vision-Language-Action Model
RotVLA models latent actions as continuous SO(n) rotations with triplet-frame supervision and flow-matching to reach 98.2% success on LIBERO and 89.6%/88.5% on RoboTwin2.0 using a 1.7B-parameter model.
-
Morphologically Equivariant Flow Matching for Bimanual Mobile Manipulation
A morphologically equivariant flow matching policy for bimanual robots enforces reflective symmetry to improve sample efficiency and enable zero-shot generalization to mirrored task configurations.
-
From Imagined Futures to Executable Actions: Mixture of Latent Actions for Robot Manipulation
MoLA infers a mixture of latent actions from generated future videos via modality-aware inverse dynamics models to improve robot manipulation policies.
-
Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models
MCF-Proto adds a motion-centric local action frame and prototype parameterization to VLA models, inducing emergent geometric structure and improved robustness from standard demonstrations alone.
-
RIO: Flexible Real-Time Robot I/O for Cross-Embodiment Robot Learning
RIO introduces a lightweight open-source framework that abstracts real-time robot I/O to support easy switching between embodiments and platforms for collecting data and deploying VLAs.
-
Offline Policy Evaluation for Manipulation Policies via Discounted Liveness Formulation
A liveness-based Bellman operator enables conservative offline policy evaluation for manipulation tasks by encoding task progression and reducing truncation bias from finite horizons.
-
Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models
Pace-and-Path Correction decomposes a quadratic cost minimization into orthogonal pace and path channels to correct chunked actions in VLA models, raising success rates by up to 28.8% in dynamic settings.
-
ALAM: Algebraically Consistent Latent Action Model for Vision-Language-Action Models
ALAM creates algebraically consistent latent action transitions from videos to act as auxiliary generative targets, raising robot policy success rates from 47.9% to 85.0% on MetaWorld MT50 and 94.1% to 98.1% on LIBERO.
-
VEGA: Visual Encoder Grounding Alignment for Spatially-Aware Vision-Language-Action Models
VEGA improves spatial reasoning in VLA models for robotics by aligning visual encoder features with 3D-supervised DINOv2 representations via a temporary projector and cosine similarity loss.
-
ECHO: Continuous Hierarchical Memory for Vision-Language-Action Models
ECHO organizes VLA experiences into a hierarchical memory tree in hyperbolic space via autoencoder and entailment constraints, delivering a 12.8% success-rate gain on LIBERO-Long over the pi0 baseline.
-
One Token Per Frame: Reconsidering Visual Bandwidth in World Models for VLA Policy
Reducing visual input to one token per frame in VLA world models maintains or improves long-horizon performance on MetaWorld, LIBERO, and real-robot tasks.
-
OA-WAM: Object-Addressable World Action Model for Robust Robot Manipulation
OA-WAM uses persistent address vectors and dynamic content vectors in object slots to enable addressable world-action prediction, improving robustness on manipulation benchmarks under scene changes.
-
Hitting Time Isomorphism for Multi-Stage Planning with Foundation Policies
A hitting-time isomorphism framework learns asymmetric Hilbert-space geometries for offline RL, yielding the IEL algorithm with identifiability proofs and improved maze navigation performance.
-
MolmoAct2: Action Reasoning Models for Real-world Deployment
MolmoAct2 delivers an open VLA model with new specialized components, datasets, and techniques that outperforms baselines on benchmarks while releasing all weights, code, and data for real-world robot use.
-
Thinking in Text and Images: Interleaved Vision--Language Reasoning Traces for Long-Horizon Robot Manipulation
A multimodal transformer generates and caches interleaved text-image traces to guide closed-loop actions, achieving 92.4% success on LIBERO-Long and 95.5% average on LIBERO.
-
Being-H0.7: A Latent World-Action Model from Egocentric Videos
Being-H0.7 adds future-aware latent reasoning to direct VLA policies via dual-branch alignment on latent queries, matching world-model benefits at VLA efficiency.
-
CodeGraphVLP: Code-as-Planner Meets Semantic-Graph State for Non-Markovian Vision-Language-Action Models
CodeGraphVLP uses a semantic-graph state and executable code planner to enable reliable long-horizon non-Markovian robot manipulation, improving task success and lowering latency over standard VLA baselines.
-
${\pi}_{0.7}$: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities
π₀.₇ is a steerable generalist robotic model that uses rich multimodal prompts including language, subgoal images, and performance metadata to achieve out-of-the-box generalization across tasks and robot bodies.
-
STRONG-VLA: Decoupled Robustness Learning for Vision-Language-Action Models under Multimodal Perturbations
STRONG-VLA uses decoupled two-stage training to improve VLA model robustness, yielding up to 16% higher task success rates under seen and unseen perturbations on the LIBERO benchmark.
-
Learning Vision-Language-Action World Models for Autonomous Driving
VLA-World improves autonomous driving by using action-guided future image generation followed by reflective reasoning over the imagined scene to refine trajectories.
-
Action Images: End-to-End Policy Learning via Multiview Video Generation
Action Images turn robot arm motions into interpretable multiview pixel videos, letting video backbones serve as zero-shot policies for end-to-end robot learning.
-
BiCoord: A Bimanual Manipulation Benchmark towards Long-Horizon Spatial-Temporal Coordination
BiCoord is a new benchmark for long-horizon tightly coordinated bimanual manipulation that includes quantitative metrics and shows existing policies like DP, RDT, Pi0 and OpenVLA-OFT struggle on such tasks.
-
VP-VLA: Visual Prompting as an Interface for Vision-Language-Action Models
VP-VLA decouples high-level reasoning from low-level control in VLA models by rendering spatial anchors as visual prompts directly in the RGB observation space, outperforming end-to-end baselines.
-
AR-VLA: True Autoregressive Action Expert for Vision-Language-Action Models
AR-VLA introduces a standalone autoregressive action expert with long-lived memory that generates context-aware continuous actions for VLAs, replacing chunk-based heads with smoother trajectories and maintained task success.
-
Evo-Memory: Benchmarking LLM Agent Test-time Learning with Self-Evolving Memory
Evo-Memory is a new benchmark for self-evolving memory in LLM agents across task streams, with baseline ExpRAG and proposed ReMem method that integrates reasoning, actions, and memory updates for continual improvement.
-
3D-VLA: A 3D Vision-Language-Action Generative World Model
3D-VLA is a new embodied foundation model that uses a 3D LLM plus aligned diffusion models to generate future images and point clouds for improved reasoning and action planning in 3D environments.
-
VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models
VoxPoser uses LLMs to compose 3D value maps via VLM interaction for model-based synthesis of robust robot trajectories on open-set language-specified manipulation tasks.
-
Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
Low-cost imprecise robots achieve 80-90% success on six fine bimanual manipulation tasks using imitation learning with a new Action Chunking with Transformers algorithm trained on only 10 minutes of demonstrations.
-
GuidedVLA: Specifying Task-Relevant Factors via Plug-and-Play Action Attention Specialization
GuidedVLA improves VLA success rates by manually supervising separate attention heads in the action decoder with auxiliary signals for task-relevant factors.
-
See What Matters: Differentiable Grid Sample Pruning for Generalizable Vision-Language-Action Model
GridS reduces visual tokens in VLA models to under 10% of the original count via task-aware differentiable resampling, delivering 76% lower FLOPs with no drop in task success rate on benchmarks and real robots.
-
Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models
Pace-and-Path Correction is a closed-form inference-time operator that decomposes a quadratic cost minimization into orthogonal pace compression and path offset channels to correct dynamics-blindness in chunked-action...
-
HarmoWAM: Harmonizing Generalizable and Precise Manipulation via Adaptive World Action Models
HarmoWAM unifies predictive and reactive control in world action models via an adaptive gating mechanism to deliver improved zero-shot generalization and precision in robotic manipulation.
-
Unified Noise Steering for Efficient Human-Guided VLA Adaptation
UniSteer unifies human corrective actions and noise-space RL for VLA adaptation by inverting actions to noise targets, raising success rates from 20% to 90% in 66 minutes across four real-world manipulation tasks.
-
ALAM: Algebraically Consistent Latent Action Model for Vision-Language-Action Models
ALAM introduces algebraic consistency regularization on latent action transitions from videos, raising VLA success rates from 47.9% to 85.0% on MetaWorld MT50 and 94.1% to 98.1% on LIBERO.
-
Retrieve-then-Steer: Online Success Memory for Test-Time Adaptation of Generative VLAs
Retrieve-then-steer stores successful observation-action segments in memory, retrieves relevant chunks, filters them, and uses an elite prior with confidence-adaptive guidance to steer a flow-matching action sampler f...
-
Retrieve-then-Steer: Online Success Memory for Test-Time Adaptation of Generative VLAs
A retrieve-then-steer method stores successful robot actions in memory and uses them to steer a frozen VLA's flow-matching sampler for better test-time reliability without parameter updates.
-
StereoPolicy: Improving Robotic Manipulation Policies via Stereo Perception
StereoPolicy fuses stereo image pairs via a Stereo Transformer on pretrained 2D encoders to boost robotic manipulation policies, showing gains over monocular, RGB-D, point cloud, and multi-view methods in simulations ...
-
Geometric Pareto Control: Riemannian Gradient Flow of Energy Function via Lie Group Homotopy
Geometric Pareto Control embeds Pareto solutions in a Lie group submanifold and navigates via Riemannian gradient flow to achieve 100% feasibility and low suboptimality in control tasks without retraining.
-
Distilling 3D Spatial Reasoning into a Lightweight Vision-Language Model with CoT
Distills 3D spatial reasoning from a 7B teacher VLM to a 2.29B student using VGGT encoder, multi-task losses, and Hidden CoT latent tokens, yielding 8.7x lower latency with 54-72% performance retention on ScanNet and ...
-
One Token Per Frame: Reconsidering Visual Bandwidth in World Models for VLA Policy
Reducing visual input to one token per frame via adaptive attention pooling and a unified flow-matching objective improves long-horizon performance in VLA policies on MetaWorld, LIBERO, and real-robot tasks.
-
One Token Per Frame: Reconsidering Visual Bandwidth in World Models for VLA Policy
Reducing visual input to one token per frame in world models for vision-language-action policies maintains long-horizon performance while improving success rates on MetaWorld, LIBERO, and real-robot tasks.
-
EggHand: A Multimodal Foundation Model for Egocentric Hand Pose Forecasting
EggHand unifies VLA action decoding with viewpoint-aware video-text encoding to forecast egocentric hand poses, achieving SOTA accuracy on EgoExo4D while remaining robust to ego-motion and controllable via language prompts.
-
AT-VLA: Adaptive Tactile Injection for Enhanced Feedback Reaction in Vision-Language-Action Models
AT-VLA introduces adaptive tactile injection and a dual-stream tactile reaction mechanism to integrate real-time tactile feedback into pretrained VLA models for contact-rich robotic manipulation.
-
HumanNet: Scaling Human-centric Video Learning to One Million Hours
HumanNet is a 1M-hour human-centric video dataset with interaction annotations that enables better vision-language-action model performance than equivalent robot data in a controlled test.
-
TriRelVLA: Triadic Relational Structure for Generalizable Embodied Manipulation
TriRelVLA introduces triadic object-hand-task relational representations and a task-grounded graph transformer with a relational bottleneck to improve generalization in robotic manipulation across scenes, objects, and tasks.
-
How Far Are VLMs from Privacy Awareness in the Physical World? An Empirical Study
VLMs show consistent deficits in identifying sensitive items in cluttered scenes, adapting to social contexts, and resolving conflicts between commands and privacy constraints in a new physical simulator benchmark.
-
How Far Are VLMs from Privacy Awareness in the Physical World? An Empirical Study
Vision-language models exhibit perceptual fragility and fail to consistently respect privacy constraints when operating in simulated physical environments, with performance declining in cluttered scenes and under conf...
-
From Pixels to Tokens: A Systematic Study of Latent Action Supervision for Vision-Language-Action Models
A unified comparison of latent action supervision strategies for VLA models reveals task-specific benefits, with image-based approaches aiding reasoning and generalization, action-based aiding motor control, and discr...
-
MolmoAct2: Action Reasoning Models for Real-world Deployment
MolmoAct2 is an open VLA model that outperforms baselines like Pi-05 on 7 benchmarks and whose backbone surpasses GPT-5 on 13 embodied-reasoning tasks through new datasets, specialized training, and architecture chang...
-
Seeing Realism from Simulation: Efficient Video Transfer for Vision-Language-Action Data Augmentation
A video transfer pipeline augments simulated VLA data into realistic videos while preserving actions, yielding consistent performance gains on robot benchmarks such as 8% on Robotwin 2.0.
-
MSACT: Multistage Spatial Alignment for Stable Low-Latency Fine Manipulation
MSACT improves localization stability and task success rates in limited-data bimanual manipulation by extracting stable 2D attention points and aligning predicted attention sequences across frames without keypoint labels.
-
Learning while Deploying: Fleet-Scale Reinforcement Learning for Generalist Robot Policies
Fleet-scale RL framework improves a single generalist VLA policy from deployment data to 95% average success on eight real-world manipulation tasks with 16 dual-arm robots.
-
LaST-R1: Reinforcing Robotic Manipulation via Adaptive Physical Latent Reasoning
LaST-R1 reaches 99.8% average success on the LIBERO benchmark using one-shot warm-up plus LAPO reinforcement learning on latent physical reasoning, with up to 44% real-world gains on complex single- and dual-arm tasks.
-
LaST-R1: Reinforcing Robotic Manipulation via Adaptive Physical Latent Reasoning
LaST-R1 introduces a RL post-training method called LAPO that optimizes latent Chain-of-Thought reasoning in vision-language-action models, yielding 99.9% success on LIBERO and up to 22.5% real-world gains.
-
PRTS: A Primitive Reasoning and Tasking System via Contrastive Representations
PRTS pretrains VLA models with contrastive goal-conditioned RL to embed goal-reachability probabilities from offline data, yielding SOTA results on robotic benchmarks especially for long-horizon and novel instructions.
-
FreqCache: Accelerating Embodied VLN Models with Adaptive Frequency-Guided Token Caching
FreqCache uses frequency domain properties to adaptively select, refresh, and budget token caches in VLN models, delivering 1.59x speedup with negligible overhead.
-
BridgeACT: Bridging Human Demonstrations to Robot Actions via Unified Tool-Target Affordances
BridgeACT learns robot manipulation from human videos alone by predicting task-relevant grasp regions and 3D motion affordances that map directly to robot controllers.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.