# smolvla

SmolVLA is a compact Vision-Language-Action model for affordable robotics.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## A$^2$-VLA: Adapting Vision-Language-Action Policies to New Tools through Agentic Affordance Discovery

DevFeed: [A$^2$-VLA: Adapting Vision-Language-Action Policies to New Tools through Agentic Affordance Discovery](<https://devfeed.tech/articles/a-2-vla-adapting-vision-language-action-policies-to-new-tools-through-agentic-affordance-discovery-145563.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25656>)

Author: Chenyuan Zhou, Chaoyi Huang, Zibo Wang, Zhiyuan Li, Yong-Lu Li

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [smolvla](<https://devfeed.tech/topics/smolvla.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [adapting](<https://devfeed.tech/tags/adapting.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [based](<https://devfeed.tech/tags/based.md>), [but](<https://devfeed.tech/tags/but.md>), [can](<https://devfeed.tech/tags/can.md>), [challenge](<https://devfeed.tech/tags/challenge.md>), [combines](<https://devfeed.tech/tags/combines.md>), [completion](<https://devfeed.tech/tags/completion.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [discovery](<https://devfeed.tech/tags/discovery.md>), [language](<https://devfeed.tech/tags/language.md>), [policies](<https://devfeed.tech/tags/policies.md>), [tools](<https://devfeed.tech/tags/tools.md>), [vision](<https://devfeed.tech/tags/vision.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>), [vla](<https://devfeed.tech/tags/vla.md>)

### AI overview

A$^2$-VLA adapts manipulation policies to unseen tools by discovering and validating task-relevant affordances. From teleoperated demonstrations with a few source tools, an embodied agent reconstructs scenes in simulation, identifies functional regions and their geometric properties, then refines affordance features and extraction skills through autonomous experiments. Validated skills are collected for reuse on unseen tools; encoded functional-region features condition a π₀.₅ backbone during training and inference. The abstract reports 86.1% success in simulated affordance-based cross-tool trajectory retargeting and a 30.7 percentage-point task-success improvement over π₀.₅ across three real-robot tool-use tasks.

### Source excerpt

arXiv:2611.25656v1 Announce Type: cross Abstract: Tools with different shapes can have similar functions, but transferring manipulation skills across tools requires understanding how tool geometry relates to the specific task and adjusting the manipulation strategy accordingly. The key challenge is to discover the common affordances that contribute to task completion across different tools and verify whether they support new tools in performing similar work. We present A$^2$-VLA, a tool-adaptation framework that combines an embodied agent with a VLA policy. It treats tool affordances as hypotheses the agent can construct, test, and revise. Given teleoperated demonstrations with a few source tools, the agent reconstructs the scenes in simulation and autonomously determines task-relevant affordance features: functional regions and their key geometric properties. It develops affordance-extraction skills and refines both the features and skills through autonomous experiments in the reconstructed scenes. Validated extraction skills accumulate in a reusable skill library for unseen tools. Finally, A$^2$-VLA uses encoded functional-region features as conditioning inputs to a $\pi_{0.5}$ backbone during training and inference. Evaluation examines affordance reliability in reconstructed scenes and real-robot task success on unseen tools. Our method achieves 86.1% success in simulated cross-tool trajectory retargeting based on affordances. On three real-robot tool-use tasks, A$^2$-VLA improves task success by 30.7 percentage points over $\pi_{0.5}$.

## Refinement vs. Replanning: Inference Compute Allocation in Robot Policies

DevFeed: [Refinement vs. Replanning: Inference Compute Allocation in Robot Policies](<https://devfeed.tech/articles/refinement-vs-replanning-inference-compute-allocation-in-robot-policies-137380.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19034>)

Author: Alba Maria Tellez Fernandez

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [robot sense of touch](<https://devfeed.tech/topics/robot-sense-of-touch.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [smolvla](<https://devfeed.tech/topics/smolvla.md>), [benchmark overfitting machine learning](<https://devfeed.tech/topics/benchmark-overfitting-machine-learning.md>)

Tags: [compute](<https://devfeed.tech/tags/compute.md>), [compute-allocation](<https://devfeed.tech/tags/compute-allocation.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [inference](<https://devfeed.tech/tags/inference.md>), [robot](<https://devfeed.tech/tags/robot.md>)

### AI overview

The case study finds that robot-policy inference compute should be evaluated across both action refinement and replanning frequency. With a frozen SmolVLA checkpoint in synchronous LIBERO simulation and chunk length fixed at K=50, reducing denoising from d=5 to d=2 at h=2 raised success from 122/160 to 139/160 and used 5.9% less arithmetic per control step. At d=2, increasing execution horizon from h=2 to h=5 cut arithmetic per attempt by 60.7%, while successes were similar (139/160 vs. 138/160); the study does not establish equivalence. Arithmetic and host-time latency also reflect different cost profiles. Results come from one checkpoint and setting, so they do not identify a universal deployment configuration.

### Source excerpt

arXiv:2611.19034v1 Announce Type: cross Abstract: A frozen chunked robot policy can spend deployment-time inference compute in two coupled ways: more work within each policy call to refine a predicted action chunk, or more frequent policy calls to replan from fresh observations. We study this allocation without updating model weights. Our evaluation procedure reports success with both nominal compute rate and complete attempt cost, selects comparisons on discovery tasks, and tests prespecified comparisons on held-out tasks. Comparisons added after inspecting the held-out results only summarize observed patterns. We demonstrate the procedure with one frozen SmolVLA checkpoint in synchronous LIBERO simulation, using denoising count as the refinement control and fixing chunk length at K=50. Arithmetic and latency characterize cost differently: at d=10, 82.1% of counted arithmetic is independent of denoising, whereas a descriptive affine fit to fixed-input host time assigns 17.7% to its intercept. At h=2, d=2 succeeds on 139/160 attempts versus 122/160 for d=5 (+10.6 percentage points; task-conditional 95% interval: +4.4 to +16.9), while using 5.9% less arithmetic per control step. Executing h=5 rather than h=2 at d=2 yields 138/160 versus 139/160 successes and 60.7% lower arithmetic per attempt; equivalence is not established. The case study shows why deployment adaptation must evaluate generation effort and feedback frequency jointly; it does not define a universal setting.

## When Backdoors Can Be Switched Off: Suppression-Aware Backdoor Attacks on Vision-Language-Action Models

DevFeed: [When Backdoors Can Be Switched Off: Suppression-Aware Backdoor Attacks on Vision-Language-Action Models](<https://devfeed.tech/articles/when-backdoors-can-be-switched-off-suppression-aware-backdoor-attacks-on-vision-language-action-models-139284.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19774>)

Author: Nur Afsa Syeda, Mohamed Elmahallawy

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [ai security](<https://devfeed.tech/topics/ai-security.md>), [smolvla](<https://devfeed.tech/topics/smolvla.md>), [Amazon robotics research](<https://devfeed.tech/topics/amazon-robotics-research.md>)

Tags: [attacks](<https://devfeed.tech/tags/attacks.md>), [backdoor](<https://devfeed.tech/tags/backdoor.md>), [conditional](<https://devfeed.tech/tags/conditional.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [detection](<https://devfeed.tech/tags/detection.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [instructions](<https://devfeed.tech/tags/instructions.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [policies](<https://devfeed.tech/tags/policies.md>), [risks](<https://devfeed.tech/tags/risks.md>), [security](<https://devfeed.tech/tags/security.md>), [smolvla](<https://devfeed.tech/tags/smolvla.md>), [trigger](<https://devfeed.tech/tags/trigger.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>)

### AI overview

The study tests VLA backdoors that activate only when multiple conditions are met and can be suppressed by an instruction. Its variants include Cross-Modal Trigger, which combines robot-state and spatially separated visual triggers; Self-Suppressing Trigger, which uses a state trigger plus [DISRUPT] and is suppressed by [RESTORE]; and Multi-Trigger, which adds a visual trigger while allowing suppression even when physical triggers persist. In SmolVLA experiments on two SO-100 datasets, Self-Suppressing and Multi-Trigger reduced Isolation Forest detection from 96-98% to 1-18% and action-space divergence by 10-14x, while reporting 63-71% attack success and perfect clean-task performance. OpenVLA experiments on LIBERO reportedly transferred the attacks to a 7B-parameter policy, evading STRIP and action-distribution-shift detection. The abstract argues conditional suppression should be included when evaluating VLA backdoor threats.

### Source excerpt

arXiv:2611.19774v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models map visual observations, robot states, and natural-language instructions to robotic actions, but their multimodal interfaces introduce new security risks. Existing VLA backdoor attacks primarily study trigger-based activation and stealthy action manipulation, leaving conditional suppression of malicious behavior underexplored. We investigate suppression-aware backdoors, which combine conjunctive trigger activation with instruction-conditioned suppression. We instantiate this design through three variants: Cross-Modal Trigger, combining robot-state and spatially separated visual triggers; Self-Suppressing Trigger, requiring a state trigger and a [DISRUPT] instruction token for activation, with [RESTORE] token suppressing the attack; and Multi-Trigger, additionally requiring a visual trigger while retaining suppression even when physical triggers persist. We evaluate these attacks using SmolVLA on two SO-100 manipulation datasets against four baselines: State Backdoor, BadVLA, TabVLA, and SilentDrift. Self-Suppressing and Multi-Trigger reduce Isolation Forest detection from 96--98% to 1--18% and action-space l divergence by 10--14X, achieving 63--71% attack success rates with perfect clean-task performance. Experiments with OpenVLA on LIBERO further show that the attacks transfer to a 7B-parameter policy, evading STRIP and action-distribution-shift detection. Our findings establish conditional suppression as an important dimension for evaluating backdoor threats in VLA policies.

## VLA Handover: Failure-Prediction-Based Switching from On-Robot to Server VLA Models

DevFeed: [VLA Handover: Failure-Prediction-Based Switching from On-Robot to Server VLA Models](<https://devfeed.tech/articles/vla-handover-failure-prediction-based-switching-from-on-robot-to-server-vla-models-123335.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.08987>)

Author: Takayuki Nishio

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>), [arXiv: Networking](<https://devfeed.tech/sources/arxiv-networking.md>)

Topics: [Isaac](<https://devfeed.tech/topics/isaac.md>), [smolvla](<https://devfeed.tech/topics/smolvla.md>)

Tags: [5g](<https://devfeed.tech/tags/5g.md>), [6g](<https://devfeed.tech/tags/6g.md>), [completion](<https://devfeed.tech/tags/completion.md>), [control](<https://devfeed.tech/tags/control.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ni](<https://devfeed.tech/tags/cs-ni.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [fail](<https://devfeed.tech/tags/fail.md>), [failure](<https://devfeed.tech/tags/failure.md>), [fewer](<https://devfeed.tech/tags/fewer.md>), [framework](<https://devfeed.tech/tags/framework.md>), [inference](<https://devfeed.tech/tags/inference.md>), [language](<https://devfeed.tech/tags/language.md>), [large](<https://devfeed.tech/tags/large.md>), [latency](<https://devfeed.tech/tags/latency.md>), [less](<https://devfeed.tech/tags/less.md>), [models](<https://devfeed.tech/tags/models.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [robot](<https://devfeed.tech/tags/robot.md>), [server](<https://devfeed.tech/tags/server.md>), [switching](<https://devfeed.tech/tags/switching.md>), [vla](<https://devfeed.tech/tags/vla.md>)

### AI overview

VLA handover uses an on-robot small VLA (S-VLA) for routine actions and calls a server-hosted large VLA (L-VLA) when the small model is likely to fail. The approach frames switching as minimizing server requests while meeting a required success probability, then uses a heuristic that retrains neither model. In simulations using SmolVLA and π₀.₅ on three LIBERO suites, it improved success by 11-22 percentage points over the S-VLA alone. Compared with server-only control, it sent 49-77% fewer requests with success 2-4 percentage points lower, and completion time increased less as network latency rose. These findings are limited to the reported simulation suites.

### Source excerpt

arXiv:2611.08987v1 Announce Type: cross Abstract: Large vision-language-action (VLA) models generalize well but are difficult to run in real time on a robot, and small VLAs that run there fail more often. Remote inference over beyond-5G and 6G networks lets a robot use a large VLA on a server, but querying the server for every action chunk limits the number of robots per server. We propose VLA handover: the robot runs a small VLA (S-VLA) and switches to a large VLA (L-VLA) on a server only when the S-VLA is likely to fail. We formulate the switching as minimizing the request cost under a required success probability and, since this problem cannot be solved directly, propose a heuristic framework that retrains neither VLA. In simulation with SmolVLA and $\pi_{0.5}$ on three of the four LIBERO suites, VLA handover raised the success rate by 11-22 percentage points over the S-VLA alone and sent 49-77% fewer requests than server-only control at a 2-4 percentage-point lower success rate, and its completion time grew less with the network latency.

## CLAP: Direct VLM-to-VLA Adaptation via Language-Action Grounding

DevFeed: [CLAP: Direct VLM-to-VLA Adaptation via Language-Action Grounding](<https://devfeed.tech/articles/clap-direct-vlm-to-vla-adaptation-via-language-action-grounding-117255.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.08974>)

Author: Yuri Ishitoya, Jeremy Siburian, Masashi Hamaya, Haruto Suzuki, Kuniaki Saito, Toshihiko Fukushima, Cristian C. Beltran-Hernandez, Mai Nishimura

Published: 2026-11-04T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [smolvla](<https://devfeed.tech/topics/smolvla.md>), [vlm](<https://devfeed.tech/topics/vlm.md>), [LLM Techniques](<https://devfeed.tech/topics/llm-techniques.md>), [post-training](<https://devfeed.tech/topics/post-training.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [grounding](<https://devfeed.tech/tags/grounding.md>), [natural-language](<https://devfeed.tech/tags/natural-language.md>), [robot](<https://devfeed.tech/tags/robot.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [scale](<https://devfeed.tech/tags/scale.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>), [vision-language-action-models](<https://devfeed.tech/tags/vision-language-action-models.md>), [vlm](<https://devfeed.tech/tags/vlm.md>)

### AI overview

CLAP addresses the mismatch that can arise when a pretrained vision-language model is adapted to predict bare numeric robot actions. It prepends each action sequence with a natural-language description, using that plan to condition precise action-token prediction without changing the backbone architecture. The abstract reports that 2B CLAP, trained for 4-5 hours, scores 90.8% on LIBERO--14.9 points above VLA-0--and is more robust to language, object, and spatial perturbations on LIBERO-PRO. On a real UR5e, it reports 60% success versus 15% for matched VLA-0. Ablations attribute gains to action-grounded language semantics. Open-weight models are released at 0.8B, 2B, and 4B; the supplied excerpt does not detail their training setup or evaluation beyond these results.

### Source excerpt

arXiv:2607.08974v2 Announce Type: replace-cross Abstract: Vision-language-action models (VLAs) inherit semantic capabilities from pretrained VLMs, yet large-scale post-training on robot data and architectural modifications can reshape the backbone so extensively that it becomes difficult to isolate what the VLM contributes to control. Directly converting pretrained VLMs into VLAs with minimal architectural change offers a more transparent path to understanding how VLM capabilities transfer across model scales. The core obstacle is output-distribution mismatch: predicting actions as bare numeric token sequences moves generation away from the VLM's pretrained language distribution, degrading the capabilities we seek to preserve. To address this, we propose CLAP (Causal Language-Action Prediction), which prepends each numeric action sequence with a natural-language action description, causally conditioning precise action-token prediction on a language-action plan without modifying the backbone architecture. With only 4-5 hours of training, 2B CLAP achieves 90.8% on LIBERO (+14.9 pt over VLA-0) and improves robustness on LIBERO-PRO under language, object, and spatial perturbations. Ablations attribute the gains to action-grounded language semantics, and on a real UR5e robot 2B CLAP reaches 60% success against 15% for matched VLA-0. We release CLAP at 0.8B, 2B, and 4B as an open-weight, multi-scale compact VLA family from a single VLM lineage, enabling controlled analysis of VLM-to-VLA capability transfer.

## Same Path, Less Time: Training-Free Fast Execution of Vision-Language-Action Models under Inference Latency

DevFeed: [Same Path, Less Time: Training-Free Fast Execution of Vision-Language-Action Models under Inference Latency](<https://devfeed.tech/articles/same-path-less-time-training-free-fast-execution-of-vision-language-action-models-under-inference-latency-114540.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.01390>)

Author: Fuyi Chen

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [smolvla](<https://devfeed.tech/topics/smolvla.md>), [vlm](<https://devfeed.tech/topics/vlm.md>), [Robot Navigation](<https://devfeed.tech/topics/robot-navigation.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [actuator](<https://devfeed.tech/tags/actuator.md>), [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [bias](<https://devfeed.tech/tags/bias.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [execution](<https://devfeed.tech/tags/execution.md>), [fast](<https://devfeed.tech/tags/fast.md>), [inference](<https://devfeed.tech/tags/inference.md>), [latency](<https://devfeed.tech/tags/latency.md>), [layer](<https://devfeed.tech/tags/layer.md>), [motion](<https://devfeed.tech/tags/motion.md>), [progress](<https://devfeed.tech/tags/progress.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [smolvla](<https://devfeed.tech/tags/smolvla.md>), [speed](<https://devfeed.tech/tags/speed.md>), [stop](<https://devfeed.tech/tags/stop.md>), [throughput](<https://devfeed.tech/tags/throughput.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>), [vision-language-action-models](<https://devfeed.tech/tags/vision-language-action-models.md>), [vla](<https://devfeed.tech/tags/vla.md>)

### AI overview

Path-preserving Time Scaling (PTS) is a training-free execution layer for frozen vision-language-action policies that addresses inference pauses and slow demonstration speeds while preserving the planned path. Its path layer deterministically decodes each replan and joins it at the progress made during inference. Its time layer sums planned increments, holds the arm at 1x while the gripper moves, and limits speed by actuator range and chunk feasibility under latency. On LIBERO with injected inference latency, PTS increased throughput over the strongest deployable baseline by 33% for pi0.5 and 104% for SmolVLA while matching or improving success rates; it was best on all six model-suite combinations and exceeded zero-latency 1x execution. Compared with asynchronous 1x execution using the same decoder, the time layer added 28% and 38%. On LIBERO-Dynamic, PTS improved throughput 38% over stop-and-wait and was comparable to asynchronous 1x execution.

### Source excerpt

arXiv:2611.01390v1 Announce Type: cross Abstract: Vision-language-action (VLA) models predict chunks of future actions, yet deployed robots waste time in two ways that are independent of model capability: they halt while the next chunk is inferred, and they move at the speed of slow, cautious demonstrations. Existing remedies remove one loss each, and stacking them fails, because asynchrony and speed-up each change the correspondence between planned progress and executed motion. We present Path-preserving Time Scaling (PTS), a training-free execution layer for frozen action-chunking policies built on one principle: the robot's motion is a time reparameterization of a reference path, and the path changes only in response to new observations. A path layer decodes every replan deterministically in one step and enters it at the progress consumed during inference. A time layer sums the planned increments, keeps the arm at 1x while the gripper moves, and caps the rate by the actuator's linear range and by chunk feasibility under latency. We show that misaligned entry and stochastic sampling add a bias and a variance to the continuation error, and derive a break-even success rate that states when speeding up pays off. On LIBERO with two frozen VLAs under injected inference latency, PTS raises task throughput over the strongest deployable baseline by 33% for pi0.5 and 104% for SmolVLA while matching or improving the success rate, is best on all six model-suite combinations, and exceeds zero-latency 1x execution. Against asynchronous 1x execution with the same one-step decoder, the time layer alone adds 28% and 38%. On LIBERO-Dynamic it improves on stop-and-wait execution by 38% and is comparable to asynchronous 1x execution.

## RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models

DevFeed: [RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models](<https://devfeed.tech/articles/repo-vla-recovery-driven-policy-optimization-for-vision-language-action-models-102958.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.09410>)

Author: Weijia Liufu, Xiaoyu Guo, Ruiyi Chen, Jingzhi Liu, Kaidong Zhang, Xiwen Liang, Jianqi Lin, Dawei Sun, Yuze Wang, Rongtao Xu, Bingqian Lin, Bowen Yang, Tongtong Cao, Bowen Peng, Dongyu Zhang, Guangrun

Published: 2026-10-28T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [contact-rich manipulation](<https://devfeed.tech/topics/contact-rich-manipulation.md>), [smolvla](<https://devfeed.tech/topics/smolvla.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [adversarial](<https://devfeed.tech/tags/adversarial.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [contact-rich-manipulation](<https://devfeed.tech/tags/contact-rich-manipulation.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [data](<https://devfeed.tech/tags/data.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [failure](<https://devfeed.tech/tags/failure.md>), [failure-recovery](<https://devfeed.tech/tags/failure-recovery.md>), [framework](<https://devfeed.tech/tags/framework.md>), [imitation](<https://devfeed.tech/tags/imitation.md>), [manipulation](<https://devfeed.tech/tags/manipulation.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [planner](<https://devfeed.tech/tags/planner.md>), [policy](<https://devfeed.tech/tags/policy.md>), [progress](<https://devfeed.tech/tags/progress.md>), [real-world](<https://devfeed.tech/tags/real-world.md>), [recovery](<https://devfeed.tech/tags/recovery.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>)

### AI overview

RePO-VLA trains vision-language-action models to recover from execution errors by learning from success, recovery, and failure trajectories. Recovery-Aware Initialization turns corrective suffixes into episodes starting from adverse states; a Progress-Aware Semantic Value Function labels useful prefixes and drift or breakdowns; and Value-Conditioned Refinement favors high-progress actions. Corrective rollouts can be planner-generated or human-collected. At deployment, a fixed high-value token biases actions toward the learned success manifold, without online failure detectors or heuristic retries. On the new Failure-Recovery Benchmark and simulated and real-world bimanual tasks, the authors report average adversarial success rising from 20% to 75%, and up to 80% in scaled real-world trials.

### Source excerpt

arXiv:2605.09410v2 Announce Type: replace-cross Abstract: Vision-Language-Action (VLA) models remain brittle in long-horizon, contact-rich manipulation because success-only imitation provides little supervision for execution drift, while failed rollouts are often discarded. We introduce RePO-VLA, a recovery-driven policy optimization framework that assigns distinct roles to success, recovery, and failure trajectories. RePO-VLA first applies Recovery-Aware Initialization (RAI), slicing each corrective suffix into a new episode so corrective actions depend on the current adverse state rather than the preceding failure. It then learns a Progress-Aware Semantic Value Function (PAS-VF), aligning spatiotemporal trajectory features with instructions and successful references. The resulting labels salvage useful failure prefixes via reliability decay, while low-value labels mark drift and terminal breakdowns, teaching differences among nominal, failed, and corrective actions. The data engine turns adverse states into planner-generated or human-collected corrective rollouts, teaching recovery to the success manifold. Value-Conditioned Refinement (VCR) trains the policy to prefer high-progress actions. At deployment, a fixed high-value token biases actions toward the learned success manifold without online failure detectors or heuristic retries. We introduce the Failure-Recovery Benchmark (FRBench), with standardized error injection and recovery-focused evaluation. Across simulated and real-world bimanual tasks, RePO-VLA improves robustness, raising adversarial success from 20% to 75% on average and up to 80% in scaled real-world trials. Code and video results are available at https://1018weijia.github.io/RePO-VLA-Web/.

## If at First You Don't Succeed, Backtrack: Lightweight Recovery for Frozen VLAs

DevFeed: [If at First You Don't Succeed, Backtrack: Lightweight Recovery for Frozen VLAs](<https://devfeed.tech/articles/if-at-first-you-don-t-succeed-backtrack-lightweight-recovery-for-frozen-vlas-99680.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.32941>)

Author: Thomas Landeg, Thierry Tambe

Published: 2026-10-26T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [smolvla](<https://devfeed.tech/topics/smolvla.md>), [Amazon robotics research](<https://devfeed.tech/topics/amazon-robotics-research.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [autoregressive](<https://devfeed.tech/tags/autoregressive.md>), [control](<https://devfeed.tech/tags/control.md>), [controller](<https://devfeed.tech/tags/controller.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [detection](<https://devfeed.tech/tags/detection.md>), [experts](<https://devfeed.tech/tags/experts.md>), [failure](<https://devfeed.tech/tags/failure.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [manipulation](<https://devfeed.tech/tags/manipulation.md>), [model](<https://devfeed.tech/tags/model.md>), [policies](<https://devfeed.tech/tags/policies.md>), [policy](<https://devfeed.tech/tags/policy.md>), [recovery](<https://devfeed.tech/tags/recovery.md>), [retry](<https://devfeed.tech/tags/retry.md>), [robotic](<https://devfeed.tech/tags/robotic.md>), [robotic-manipulation](<https://devfeed.tech/tags/robotic-manipulation.md>), [training](<https://devfeed.tech/tags/training.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>), [vla](<https://devfeed.tech/tags/vla.md>)

### AI overview

Backtrack adds a small failure-prediction head to frozen vision-language-action policies and uses it to trigger recovery: when predicted failure exceeds a threshold, a controller pauses the policy, retraces recent arm motion, then lets the unchanged policy retry from an earlier configuration. The head uses existing model representations, is trained on outcome-labeled rollouts, and adds 0.48-0.88% to policy parameters. On LIBERO-Long, it reduced failure rates by 19-48% across three tested VLAs, raising success by up to 7.4 percentage points; effects were smaller on shorter suites with 85-99% baseline success. All 12 tested SmolVLA controller configurations improved on the unassisted policy on MetaWorld.

### Source excerpt

arXiv:2610.32941v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models perform well on robotic manipulation, yet remain brittle once their own actions carry them into states their training data barely covers. Recovering from such failures usually requires recovery demonstrations, policy retraining, or a separate model at inference time. We introduce Backtrack, a lightweight failure detection and recovery mechanism for frozen VLAs that needs none of these. A small failure prediction head, trained only on outcome-labeled rollouts of the frozen policy, reads the VLA's existing internal representations to estimate whether the current episode will fail. When the predicted failure probability exceeds a threshold, a controller pauses the policy, retraces the arm's recently achieved motion, and hands control back to the unchanged policy so it can retry from an earlier configuration. The head adds only 0.48-0.88% to the policy's parameter count. We evaluate Backtrack on LIBERO with three VLAs spanning two action-generation families: SmolVLA and $\pi$0.5, which use flow-matching action experts, and the autoregressive $\pi$0-FAST. On LIBERO-Long, the suite of long-horizon multi-step tasks and the one on which all three policies fail most often, Backtrack reduces each policy's failure rate by 19-48%, raising success by up to 7.4 percentage points; on the three shorter suites, where control success is already 85-99%, its effect is smaller. On MetaWorld, all twelve controller configurations tested with SmolVLA improve on the unassisted policy.

## Let It Be Simple: One-Step Action Generation for Vision-Language-Action Models

DevFeed: [Let It Be Simple: One-Step Action Generation for Vision-Language-Action Models](<https://devfeed.tech/articles/let-it-be-simple-one-step-action-generation-for-vision-language-action-models-94727.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2606.05737>)

Author: Yitong Chen, Shiduo Zhang, Jingjing Gong, Xipeng Qiu

Published: 2026-10-22T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [smolvla](<https://devfeed.tech/topics/smolvla.md>), [Generative AI](<https://devfeed.tech/topics/generative-ai.md>)

Tags: [architectures](<https://devfeed.tech/tags/architectures.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [condition](<https://devfeed.tech/tags/condition.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generative](<https://devfeed.tech/tags/generative.md>), [image](<https://devfeed.tech/tags/image.md>), [image-generation](<https://devfeed.tech/tags/image-generation.md>), [models](<https://devfeed.tech/tags/models.md>), [noise](<https://devfeed.tech/tags/noise.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [source](<https://devfeed.tech/tags/source.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [text-to-image](<https://devfeed.tech/tags/text-to-image.md>), [velocity](<https://devfeed.tech/tags/velocity.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>), [vision-language-action-models](<https://devfeed.tech/tags/vision-language-action-models.md>)

### AI overview

The paper reports that VLA action generation can often work in one step with standard Flow Matching: biasing training timesteps toward the noise endpoint yields strong policies across LIBERO variants and real-world robot tasks, while single-pass decoding in several open-source models often performs comparably to iterative decoding on LIBERO, CALVIN, and SimplerEnv. The authors attribute this to rich observations constraining compact action targets. Supporting this explanation, their measured irreducible velocity loss decreases toward the noise endpoint for VLA, unlike the typically U-shaped image-generation profile, and the learned action map has a smaller largest Jacobian singular value with respect to initial noise than the image map. The supplied text is an abstract, so it does not give detailed methods or per-task results.

### Source excerpt

arXiv:2606.05737v3 Announce Type: replace-cross Abstract: One-step generation from standard diffusion or Flow-Matching training is rarely effective for text-to-image or video generation without distillation or specialized one-step objectives. We find that VLA action generation behaves strikingly differently. Biasing only the training timestep distribution toward the noise endpoint yields strong one-step policies across LIBERO, LIBERO-Plus, LIBERO-Pro, and real-world robot tasks. Moreover, for open-source VLA models with different architectures, simply replacing iterative decoding with a single forward pass often performs comparably to the original multi-step decoding on benchmarks such as LIBERO, CALVIN, and SimplerEnv. We attribute this difference to the condition-target structure of VLA generation, which is closer to image-to-text than text-to-image: rich observations strongly constrain a comparatively compact target. We formalize this intuition through the irreducible velocity loss $R_v(t,c)$ of standard Flow Matching, which becomes much smaller toward the noise endpoint for VLA, whereas image-generation models usually exhibit a U-shaped profile. The geometry induced by the learned generative map shows the same contrast: the Jacobian of this map with respect to the initial noise has a substantially smaller largest singular value for action generation than for image generation. Together, these results suggest that VLA's condition-target structure makes one-step generation unusually accessible to standard Flow Matching. Code and training configurations for this paper are available at https://github.com/bondtesty/let-it-be-simple.

## Post-hoc Epistemic Uncertainty for Vision-Language-Action Policies

DevFeed: [Post-hoc Epistemic Uncertainty for Vision-Language-Action Policies](<https://devfeed.tech/articles/post-hoc-epistemic-uncertainty-for-vision-language-action-policies-94981.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.29730>)

Author: Johannes Keller, Albrecht M. Blo{\\ss}e, Adrian Lindenmeyer, Thomas Neumuth, Daniel Schneider

Published: 2026-10-22T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [smolvla](<https://devfeed.tech/topics/smolvla.md>), [Amazon robotics research](<https://devfeed.tech/topics/amazon-robotics-research.md>)

Tags: [acquisition](<https://devfeed.tech/tags/acquisition.md>), [adaptive](<https://devfeed.tech/tags/adaptive.md>), [camera](<https://devfeed.tech/tags/camera.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [decoder](<https://devfeed.tech/tags/decoder.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [policies](<https://devfeed.tech/tags/policies.md>), [recognition](<https://devfeed.tech/tags/recognition.md>), [robot](<https://devfeed.tech/tags/robot.md>), [setup](<https://devfeed.tech/tags/setup.md>), [smolvla](<https://devfeed.tech/tags/smolvla.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [state](<https://devfeed.tech/tags/state.md>), [supervisor](<https://devfeed.tech/tags/supervisor.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [time](<https://devfeed.tech/tags/time.md>), [training](<https://devfeed.tech/tags/training.md>), [uncertainty](<https://devfeed.tech/tags/uncertainty.md>), [variance](<https://devfeed.tech/tags/variance.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>)

### AI overview

The paper proposes post-hoc kernel-variance heads that estimate epistemic uncertainty from a single VLA policy's embeddings, using a Gaussian process conditioned on training demonstrations. Fitting requires neither regression targets nor gradient updates and leaves the policy unchanged. In SmolVLA experiments on LIBERO-10, the heads performed competitively with two-policy ensemble scores for task-success correlation, active-learning demonstration acquisition, and referral to a stronger supervisor. The vision-language backbone head cut uncertainty-scoring time by up to about 50-fold versus ensemble scoring and reduced setup costs. The authors caution that LIBERO-10's standard protocol mainly rewards recognizing task coverage differences, while execution variability obscures within-task rankings. Controlled perturbations complement that evaluation: modality-specific heads detected unfamiliar camera, instruction, or state inputs, and fused heads detected unfamiliar combinations. With a supervisor proxy trained on those perturbation families, the decoder head improved within-task referral over ensemble disagreement.

### Source excerpt

arXiv:2610.29730v1 Announce Type: cross Abstract: Vision-language-action (VLA) policies need to recognize gaps in their training knowledge to request useful demonstrations or assistance in new situations, especially when failures are costly. Disagreement within an ensemble of independently trained policies can guide these decisions, but requires training and running multiple policies. We propose post-hoc kernel-variance heads that estimate epistemic uncertainty from the embeddings of a single policy. Each head computes the conditional variance of a Gaussian process conditioned on embeddings from the training demonstrations. Fitting requires no regression targets or gradient updates and leaves the policy unchanged. In experiments with SmolVLA on LIBERO-10, the kernel variance heads perform competitively with two-policy ensemble scores in task-level correlation with task success, in demonstration acquisition during active learning, and in referral to a stronger supervisor policy as a proxy for human assistance. For SmolVLA, the head on the vision-language backbone reduces uncertainty-scoring time by up to approximately a factor of 50 compared with ensemble scores, while also substantially reducing setup costs. We also show that the established evaluation protocol on LIBERO-10 primarily rewards recognition of coverage differences between tasks, while execution variability weakens assessment of within-task rankings. Thus, controlled perturbations provide a complementary evaluation: modality-specific heads identify unfamiliar camera, instruction, or state inputs, while fused heads can detect unfamiliar combinations of individually familiar inputs. With a supervisor proxy trained on the evaluated perturbation families, the decoder head improves within-task referral over ensemble disagreement. These findings show that useful uncertainty estimates for adaptive robot learning can be obtained from a single policy.

## Meet Kyutai REKI: The 2.4B VLA Model Bringing Efficient Memory to Robot Manipulation

DevFeed: [Meet Kyutai REKI: The 2.4B VLA Model Bringing Efficient Memory to Robot Manipulation](<https://devfeed.tech/articles/meet-kyutai-reki-the-2-4b-vla-model-bringing-efficient-memory-to-robot-manipulation-89435.md>)

Original publisher: [Read original article](<https://www.marktechpost.com/2026/10/17/meet-kyutai-reki-the-2-4b-vla-model-bringing-efficient-memory-to-robot-manipulation/>)

Author: Asif Razzaq

Published: 2026-10-17T20:22:11Z

Content type: news

Language: en

Sources: [MarkTechPost](<https://devfeed.tech/sources/marktechpost.md>)

Topics: [Robot Manipulation](<https://devfeed.tech/topics/robot-manipulation.md>), [robot sense of touch](<https://devfeed.tech/topics/robot-sense-of-touch.md>), [smolvla](<https://devfeed.tech/topics/smolvla.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [model architecture](<https://devfeed.tech/topics/model-architecture.md>), [NUMA](<https://devfeed.tech/topics/numa.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-paper-summary](<https://devfeed.tech/tags/ai-paper-summary.md>), [ai-shorts](<https://devfeed.tech/tags/ai-shorts.md>), [applications](<https://devfeed.tech/tags/applications.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [control](<https://devfeed.tech/tags/control.md>), [data](<https://devfeed.tech/tags/data.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [editors-pick](<https://devfeed.tech/tags/editors-pick.md>), [embedding-model](<https://devfeed.tech/tags/embedding-model.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-memory](<https://devfeed.tech/tags/gpu-memory.md>), [harness](<https://devfeed.tech/tags/harness.md>), [kyutai](<https://devfeed.tech/tags/kyutai.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [manipulation](<https://devfeed.tech/tags/manipulation.md>), [models](<https://devfeed.tech/tags/models.md>), [new-releases](<https://devfeed.tech/tags/new-releases.md>), [physical-ai](<https://devfeed.tech/tags/physical-ai.md>), [robot](<https://devfeed.tech/tags/robot.md>), [robot-manipulation](<https://devfeed.tech/tags/robot-manipulation.md>), [robotics](<https://devfeed.tech/tags/robotics.md>), [staff](<https://devfeed.tech/tags/staff.md>), [tech-news](<https://devfeed.tech/tags/tech-news.md>), [technology](<https://devfeed.tech/tags/technology.md>), [vision](<https://devfeed.tech/tags/vision.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>), [vision-language-model](<https://devfeed.tech/tags/vision-language-model.md>), [vla](<https://devfeed.tech/tags/vla.md>)

### AI overview

Kyutai, Inria, and École des Ponts released REKI, a 2.4-billion-parameter vision-language-action model for robot manipulation, pretrained on public data. Its action expert uses cached register tokens to retain recent visual context; the article reports that this adds 0.1 ms per action chunk and reduces oscillating episodes on RoboTwin from 18.6% to 4.1%. REKI scores 90.4% on LIBERO-Plus, while results vary across other benchmarks.

### Source excerpt

Kyutai, with Inria and École des Ponts, has released REKI, a 2.4B vision-language-action model (VLA) for robot manipulation. The REKI vision-language-action model is pretrained only on public data. It also carries a cheap memory of its recent past. On three out-of-distribution benchmarks, only models about twice its size beat it. All weights are live on [...] The post Meet Kyutai REKI: The 2.4B VLA Model Bringing Efficient Memory to Robot Manipulation appeared first on MarkTechPost.

## MolmoAct 2 shows what open models can unlock for robotics

DevFeed: [MolmoAct 2 shows what open models can unlock for robotics](<https://devfeed.tech/articles/molmoact-2-shows-what-open-models-can-unlock-for-robotics-82652.md>)

Original publisher: [Read original article](<https://allenai.org/blog/pham-molmoact2-testimonial>)

Published: 2026-07-08T08:00:00Z

Content type: article

Language: en

Sources: [Ai2 Blog](<https://devfeed.tech/sources/ai2-blog.md>)

Topics: [contact-rich manipulation](<https://devfeed.tech/topics/contact-rich-manipulation.md>), [smolvla](<https://devfeed.tech/topics/smolvla.md>), [Open Source Models & Datasets](<https://devfeed.tech/topics/open-source-models-datasets.md>), [livekit](<https://devfeed.tech/topics/livekit.md>)

Tags: [embodied-ai](<https://devfeed.tech/tags/embodied-ai.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [github](<https://devfeed.tech/tags/github.md>), [hugging-face](<https://devfeed.tech/tags/hugging-face.md>), [robotics](<https://devfeed.tech/tags/robotics.md>)

### AI overview

Robotics software engineer Binh Pham used MolmoAct 2 in a voice-controlled robot that won an embodied AI hackathon. He describes the model's object recognition, spatial awareness, and ability to perform tasks it was not specifically trained for. The article also highlights the open release's model weights, data, training code, and fine-tuning scripts.

### Source excerpt

Robotics engineer Binh Pham used MolmoAct 2 to build a voice-controlled robot that won South Park Commons' embodied AI hackathon.