{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,1]],"date-time":"2026-10-01T06:00:49Z","timestamp":1790834449891,"version":"4.1.0"},"reference-count":34,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,10,1]],"date-time":"2026-10-01T00:00:00Z","timestamp":1790812800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Neurorobot."],"abstract":"<jats:p>\n                    Vision-Language-Action (VLA) models offer a unified interface for language-conditioned manipulation. This paper reports a negative result: a low-cost post-training recipe for SmolVLA with LoRA, a temporal smoothness regularizer, and chunk-aware auxiliary supervision from action-variation pseudo labels does not improve manipulation success under a 30,000-step budget (approximately 1\/67 of the original 2,000,000-step SmolVLA recipe). Because the auxiliary classifier is used only during training and the runtime chunk length remains fixed at\n                    <jats:italic>k<\/jats:italic>\n                    = 8, we denote the configuration SmolVLA+LoRA+S&amp;amp;C (smoothness and chunk-aware auxiliary losses) rather than \u201cadaptive.\u201d We evaluate four LIBERO suites with a unified protocol of 500 episodes per cell (50 episodes \u00d7 10 task variants). Over three training seeds, the matched fixed-chunk\n                    <jats:italic>k<\/jats:italic>\n                    = 8 baseline reaches a four-suite macro success rate of 10.9%\u00b11.1, whereas SmolVLA+LoRA+S&amp;amp;C reaches 3.5%\u00b11.4; the pre-specified primary comparison shows significant deficits on LIBERO-Spatial (\u221216.4 percentage points, exact sign-flip permutation\n                    <jats:italic>p<\/jats:italic>\n                    = 0.004) and LIBERO-Object (\u221217.5 pp,\n                    <jats:italic>p<\/jats:italic>\n                    = 0.002), with matched-pairs rank-biserial effect sizes of \u22120.96 and \u22121.0. Matched component ablations isolate the smoothness loss as the harmful term (smoothness-only collapses performance; chunk-classifier-only approximately matches the baseline), and executed-action logs show the regularizer\n                    <jats:italic>triples<\/jats:italic>\n                    trajectory jerk (1.69 vs. 0.59 RMS)\u2014the opposite of its intent. A training-free inference-time temporal-ensembling baseline, by contrast, more than doubles success on the two suites where the baseline is competent (Spatial 36.2 vs. 19.0%; Object 56.2 vs. 24.0%) at a 2.3 \u00d7 step-rate cost, and a 10\u2013100K budget sweep shows the S&amp;amp;C deficit never closes. We conclude that, for the tested configuration, budgets, and benchmark, the added action-side losses are not competitive with a well-tuned fixed chunk, and inference-time ensembling is the stronger use of the same compute.\n                  <\/jats:p>","DOI":"10.3389\/fnbot.2026.1890245","type":"journal-article","created":{"date-parts":[[2026,10,1]],"date-time":"2026-10-01T05:42:39Z","timestamp":1790833359000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Low-cost fine-tuning of SmolVLA with smoothness regularization and chunk-aware auxiliary supervision for language-conditioned manipulation"],"prefix":"10.3389","volume":"20","author":[{"given":"Chenkai","family":"He","sequence":"first","affiliation":[{"name":"Tianjin University of Science and Technology","place":["Tianjin, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,10,1]]},"reference":[{"key":"B1","article-title":"RoboCat: a self-improving generalist agent for robotic manipulation","author":"Bousmalis","year":"2024","journal-title":"Trans. Mach. Learn. Res"},{"key":"B2","article-title":"\u201cRT-2: vision-language-action models transfer web knowledge to robotic control,\u201d","author":"Brohan","year":"","journal-title":"Proceedings of the 7th conference on robot learning (CoRL)"},{"key":"B3","doi-asserted-by":"publisher","DOI":"10.15607\/RSS.2023.XIX.025","article-title":"\u201cRT-1: robotics transformer for real-world control at scale,\u201d","author":"Brohan","year":"","journal-title":"Robotics: science and systems (RSS)"},{"key":"B4","author":"Chen","year":"2021"},{"key":"B5","doi-asserted-by":"publisher","DOI":"10.15607\/RSS.2023.XIX.026","article-title":"\u201cDiffusion policy: visuomotor policy learning via action diffusion,\u201d","author":"Chi","year":"2023","journal-title":"Proceedings of robotics: science and systems"},{"key":"B6","first-page":"1","article-title":"Statistical comparisons of classifiers over multiple data sets","volume":"7","author":"Dem\u0161ar","year":"2006","journal-title":"J. Mach. Learn. Res"},{"key":"B7","article-title":"\u201cPaLM-E: an embodied multimodal language model,\u201d","author":"Driess","year":"2023","journal-title":"Proceedings of the 40th international conference on machine learning (ICML)"},{"key":"B8","doi-asserted-by":"publisher","first-page":"675","DOI":"10.1080\/01621459.1937.10503522","article-title":"The use of ranks to avoid the assumption of normality implicit in the analysis of variance","volume":"32","author":"Friedman","year":"1937","journal-title":"J. Am. Stat. Assoc"},{"key":"B9","doi-asserted-by":"publisher","DOI":"10.1109\/IROS.2013.6696514","article-title":"\u201cUnified temporal and spatial calibration for multi-sensor systems,\u201d","author":"Furgale","year":"2013","journal-title":"IEEE\/RSJ international conference on intelligent robots and systems (IROS)"},{"key":"B10","article-title":"\u201cLoRA: low-rank adaptation of large language models,\u201d","author":"Hu","year":"2022","journal-title":"International conference on learning representations (ICLR)"},{"key":"B11","doi-asserted-by":"publisher","first-page":"3019","DOI":"10.1109\/LRA.2020.2974707","article-title":"RLBench: the robot learning benchmark & learning environment","volume":"5","author":"James","year":"2020","journal-title":"IEEE Robot. Autom. Lett"},{"key":"B12","article-title":"\u201cBC-Z: zero-shot task generalization with robotic imitation learning,\u201d","author":"Jang","year":"2021","journal-title":"Proceedings of the 5th conference on robot learning (CoRL)"},{"key":"B13","article-title":"\u201cVIMA: general robot manipulation with multimodal prompts,\u201d","author":"Jiang","year":"2023","journal-title":"Proceedings of the 40th international conference on machine learning (ICML)"},{"key":"B14","article-title":"\u201cDROID: a large-scale in-the-wild robot manipulation dataset,\u201d","author":"Khazatsky","year":"2024","journal-title":"Robotics: science and systems (RSS)"},{"key":"B15","doi-asserted-by":"publisher","first-page":"76904","DOI":"10.1109\/ACCESS.2023.3298055","article-title":"Automatic extrinsic calibration of a camera and a 2D LiDAR with point-line correspondences","volume":"11","author":"Kim","year":"2023","journal-title":"IEEE Access"},{"key":"B16","article-title":"\u201cOpenVLA: an open-source vision-language-action model,\u201d","author":"Kim","year":"2024","journal-title":"Proceedings of the 8th conference on robot learning (CoRL)"},{"key":"B17","doi-asserted-by":"publisher","DOI":"10.52202\/075280-1939","article-title":"\u201cLIBERO: benchmarking knowledge transfer for lifelong robot learning,\u201d","author":"Liu","year":"2023","journal-title":"Advances in neural information processing systems (NeurIPS), datasets and benchmarks track"},{"key":"B18","doi-asserted-by":"publisher","DOI":"10.15607\/RSS.2021.XVII.047","article-title":"\u201cLanguage conditioned imitation learning over unstructured data,\u201d","author":"Lynch","year":"2021","journal-title":"Proceedings of robotics: science and systems XVII"},{"key":"B19","first-page":"1678","article-title":"\u201cWhat matters in learning from offline human demonstrations for robot manipulation,\u201d","volume-title":"Proceedings of the 5th conference on Robot learning","author":"Mandlekar","year":"2022"},{"key":"B20","article-title":"\u201cROBOTURK: a crowdsourcing platform for robotic skill learning through imitation,\u201d","author":"Mandlekar","year":"2018","journal-title":"Proceedings of the 2nd conference on robot learning (CoRL)"},{"key":"B21","article-title":"\u201cR3M: a universal visual representation for robot manipulation,\u201d","author":"Nair","year":"2022","journal-title":"Proceedings of the 6th conference on robot learning (CoRL)"},{"key":"B22","article-title":"\u201cOcto: an open-source generalist robot policy,\u201d","author":"Octo Model Team","year":"2024","journal-title":"Robotics: science and systems (RSS)"},{"key":"B23","article-title":"\u201cOpen X-embodiment: robotic learning datasets and RT-X models,\u201d","author":"Padalkar","year":"2024","journal-title":"IEEE international conference on robotics and automation (ICRA)"},{"key":"B24","doi-asserted-by":"publisher","first-page":"634","DOI":"10.1126\/science.abj5259","article-title":"Neurorobotics for neurorehabilitation","volume":"373","author":"Raspopovic","year":"2021","journal-title":"Science"},{"key":"B25","article-title":"A generalist agent","author":"Reed","year":"2022","journal-title":"Trans. Mach. Learn. Res"},{"key":"B26","doi-asserted-by":"publisher","DOI":"10.52202\/068431-1668","article-title":"\u201cBehavior transformers: Cloning k modes with one stone,\u201d","author":"Shafiullah","year":"2022","journal-title":"Advances in neural information processing systems (NeurIPS)"},{"key":"B27","article-title":"\u201cCLIPort: what and where pathways for robotic manipulation,\u201d","author":"Shridhar","year":"2021","journal-title":"Proceedings of the 5th conference on robot learning (CoRL)"},{"key":"B28","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2506.01844","article-title":"SmolVLA: a vision-language-action model for affordable and efficient robotics","author":"Shukor","year":"2025"},{"key":"B29","doi-asserted-by":"publisher","first-page":"104062","DOI":"10.1016\/j.inffus.2025.104062","article-title":"Multimodal fusion with vision-language-action models for robotic manipulation: a systematic review","volume":"129","author":"Ud Din","year":"2026","journal-title":"Inf. Fusion"},{"key":"B30","article-title":"\u201cBridgeData V2: a dataset for robot learning at scale,\u201d","author":"Walke","year":"2023","journal-title":"Proceedings of the 7th conference on robot learning (CoRL)"},{"key":"B31","doi-asserted-by":"publisher","first-page":"80","DOI":"10.2307\/3001968","article-title":"Individual comparisons by ranking methods","volume":"1","author":"Wilcoxon","year":"1945","journal-title":"Biometrics Bull"},{"key":"B32","article-title":"\u201cTransporter networks: rearranging the visual world for robotic manipulation,\u201d","author":"Zeng","year":"2020","journal-title":"Proceedings of the 4th conference on robot learning (CoRL)"},{"key":"B33","doi-asserted-by":"publisher","DOI":"10.15607\/RSS.2023.XIX.016","article-title":"\u201cLearning fine-grained bimanual manipulation with low-cost hardware,\u201d","author":"Zhao","year":"2023","journal-title":"Robotics: science and systems (RSS)"},{"key":"B34","doi-asserted-by":"publisher","first-page":"15635","DOI":"10.1021\/acsami.6c00064","article-title":"A bioinspired multifunctional adhesive-tactile E-Skin enabled for adaptive grasping and slip detection","volume":"18","author":"Zhu","year":"2026","journal-title":"ACS Appl. Mater. Interfaces"}],"container-title":["Frontiers in Neurorobotics"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fnbot.2026.1890245\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,1]],"date-time":"2026-10-01T05:42:43Z","timestamp":1790833363000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fnbot.2026.1890245\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,10,1]]},"references-count":34,"alternative-id":["10.3389\/fnbot.2026.1890245"],"URL":"https:\/\/doi.org\/10.3389\/fnbot.2026.1890245","relation":{},"ISSN":["1662-5218"],"issn-type":[{"value":"1662-5218","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,10,1]]},"article-number":"1890245"}}