{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,12]],"date-time":"2026-08-12T05:27:43Z","timestamp":1786512463609,"version":"build-2736575974"},"reference-count":42,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,8,12]],"date-time":"2026-08-12T00:00:00Z","timestamp":1786492800000},"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. Robot. AI"],"abstract":"<jats:p>Imitation learning enables robots to acquire complex manipulation skills from human demonstrations, but current methods rely solely on low-level sensorimotor data while ignoring the rich semantic knowledge humans naturally possess about tasks. We present ConceptACT, an extension of Action Chunking with Transformers that leverages episode-level semantic concept annotations during training to improve learning efficiency. Unlike language-conditioned approaches that require semantic input at deployment, ConceptACT uses human-provided concepts (object properties, spatial relationships, task constraints) exclusively during demonstration collection, adding minimal annotation burden. We integrate concepts using a modified transformer architecture in which the final encoder layer implements concept-aware cross-attention, supervised to align with human annotations. Through experiments on two robotic manipulation tasks with logical constraints, we demonstrate that ConceptACT converges faster and achieves superior sample efficiency compared to standard ACT. Crucially, we show that architectural integration through attention mechanisms significantly outperforms naive auxiliary prediction losses or language-conditioned models. These results indicate that properly integrated semantic supervision provides useful inductive biases for manipulation tasks.<\/jats:p>","DOI":"10.3389\/frobt.2026.1865290","type":"journal-article","created":{"date-parts":[[2026,8,12]],"date-time":"2026-08-12T04:37:57Z","timestamp":1786509477000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["ConceptACT: episode-level concepts for sample-efficient robotic imitation learning"],"prefix":"10.3389","volume":"13","author":[{"given":"Jakob","family":"Karalus","sequence":"first","affiliation":[{"name":"Institute of Artificial Intelligence, Ulm University","place":["Ulm, Germany"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Friedhelm","family":"Schwenker","sequence":"additional","affiliation":[{"name":"Institute of Neuroinformatics, Ulm University","place":["Ulm, Germany"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,8,12]]},"reference":[{"key":"B1","first-page":"29304","article-title":"Deep reinforcement learning at the edge of the statistical precipice","volume":"34","author":"Agarwal","year":"2021","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"B2","first-page":"7786","article-title":"Towards robust interpretability with self-explaining neural networks","volume":"31","author":"Alvarez Melis","year":"2018","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"B3","doi-asserted-by":"publisher","first-page":"469","DOI":"10.1016\/j.robot.2008.10.024","article-title":"A survey of robot learning from demonstration","volume":"57","author":"Argall","year":"2009","journal-title":"Robotics Autonomous Systems"},{"key":"B4","article-title":"Rt-1: robotics transformer for real-world control at scale","author":"Brohan","year":"2022"},{"key":"B5","first-page":"8928","article-title":"This looks like that: deep learning for interpretable image recognition","volume":"32","author":"Chen","year":"2019","journal-title":"Adv. Neural Inf. Process. Syst"},{"key":"B6","doi-asserted-by":"publisher","first-page":"772","DOI":"10.1038\/s42256-020-00265-z","article-title":"Concept whitening for interpretable image recognition","volume":"2","author":"Chen","year":"2020","journal-title":"Nat. Mach. Intell."},{"key":"B7","doi-asserted-by":"publisher","first-page":"1684","DOI":"10.1177\/02783649241273668","article-title":"Diffusion policy: visuomotor policy learning via action diffusion","volume":"44","author":"Chi","year":"2025","journal-title":"Int. J. Robotics Res."},{"key":"B8","doi-asserted-by":"crossref","first-page":"2592","DOI":"10.1109\/ICRA.2015.7139548","article-title":"High-level learning from demonstration with conceptual spaces and subspace clustering","volume-title":"2015 IEEE International Conference on Robotics and Automation (ICRA)","author":"Cubek","year":"2015"},{"key":"B9","doi-asserted-by":"crossref","first-page":"67156","DOI":"10.52202\/075280-2935","article-title":"State2explanation: concept-based explanations to benefit agent learning and user understanding","volume":"36","author":"Das","year":"2023","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"B10","article-title":"An image is worth 16x16 words: transformers for image recognition at scale","volume-title":"International Conference on Learning Representations","author":"Dosovitskiy","year":"2021"},{"key":"B11","doi-asserted-by":"publisher","first-page":"701","DOI":"10.1177\/02783649241281508","article-title":"Foundation models in robotics: applications, challenges, and the future","volume":"44","author":"Firoozi","year":"2025","journal-title":"Int. J. Robotics Res."},{"key":"B12","doi-asserted-by":"publisher","first-page":"516","DOI":"10.1109\/JPROC.2024.3435012","article-title":"When multitask learning meets partial supervision: a computer vision review","volume":"112","author":"Fontana","year":"2024","journal-title":"Proc. IEEE"},{"key":"B13","first-page":"21885","article-title":"Widening the pipeline in human-guided reinforcement learning with explanation and context-aware data augmentation","volume":"34","author":"Guan","year":"2021","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"B14","article-title":"Entity-centric reinforcement learning for object manipulation from pixels","volume-title":"The Twelfth International Conference on Learning Representations ICRL","author":"Haramati","year":"2024"},{"key":"B15","first-page":"770","article-title":"Deep residual learning for image recognition","volume-title":"Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition","author":"He","year":"2016"},{"key":"B16","doi-asserted-by":"publisher","first-page":"1863","DOI":"10.1177\/02783649241304789","article-title":"Furniturebench: reproducible real-world benchmark for long-horizon complex manipulation","volume":"44","author":"Heo","year":"2025","journal-title":"Int. J. Robotics Res."},{"key":"B17","doi-asserted-by":"crossref","first-page":"2288","DOI":"10.1109\/ICRA57147.2024.10610505","article-title":"Polite: preferences combined with highlights in reinforcement learning","volume-title":"2024 IEEE International Conference on Robotics and Automation (ICRA)","author":"Holk","year":"2024"},{"key":"B18","article-title":"Learning from demonstration with weakly supervised disentanglement","volume-title":"Ninth International Conference on Learning Representations 2021","author":"Hristov","year":"2021"},{"key":"B19","doi-asserted-by":"crossref","first-page":"362","DOI":"10.1109\/ICDL53763.2022.9962222","article-title":"Accelerating the learning of tamer with counterfactual explanations","volume-title":"2022 IEEE International Conference on Development and Learning (ICDL)","author":"Karalus","year":"2022"},{"key":"B20","doi-asserted-by":"publisher","first-page":"125759","DOI":"10.1109\/ACCESS.2025.3588936","article-title":"Tempo: timestep explanations for modeling preferences in online preference-based rl","volume":"13","author":"Karalus","year":"2025","journal-title":"IEEE Access"},{"key":"B21","doi-asserted-by":"crossref","DOI":"10.15607\/RSS.2023.XIX.032","article-title":"Language-driven representation learning for robotics","volume-title":"Robotics: Science and Systems (RSS)","author":"Karamcheti","year":"2023"},{"key":"B22","first-page":"2668","article-title":"Interpretability beyond feature attribution: quantitative testing with concept activation vectors (tcav)","volume-title":"International Conference on Machine Learning","author":"Kim","year":"2018"},{"key":"B23","first-page":"2679","article-title":"Openvla: an open-source vision-language-action model","volume-title":"Conference on Robot Learning","author":"Kim","year":"2025"},{"key":"B24","article-title":"Auto-encoding variational bayes","volume-title":"International Conference on Learning Representations","author":"Kingma","year":"2014"},{"key":"B25","volume-title":"Standard Open SO-100 & SO-101 Arms","author":"Knight","year":"2024"},{"key":"B26","doi-asserted-by":"crossref","first-page":"9","DOI":"10.1145\/1597735.1597738","article-title":"Interactively shaping agents via human reinforcement: the tamer framework","volume-title":"Proceedings of the Fifth International Conference on Knowledge Capture","author":"Knox","year":"2009"},{"key":"B27","first-page":"5338","article-title":"Concept bottleneck models","volume-title":"International Conference on Machine Learning","author":"Koh","year":"2020"},{"key":"B28","first-page":"309","article-title":"Structure preserving stain normalization of histopathology images using self supervised semantic guidance","volume-title":"Medical Image Computing and Computer Assisted Intervention\u2013MICCAI 2020","author":"Mahapatra","year":"2020"},{"key":"B29","first-page":"1678","article-title":"What matters in learning from offline human demonstrations for robot manipulation","volume-title":"Conference on Robot Learning","author":"Mandlekar","year":"2022"},{"key":"B30","doi-asserted-by":"publisher","first-page":"3942","DOI":"10.1609\/aaai.v32i1.11671","article-title":"Film: visual reasoning with a general conditioning layer","volume":"32","author":"Perez","year":"2018","journal-title":"Proc. AAAI Conf. Artif. Intell."},{"key":"B31","article-title":"Attention-based interpretability with concept transformers","volume-title":"International Conference on Learning Representations","author":"Rigotti","year":"2022"},{"key":"B32","doi-asserted-by":"publisher","first-page":"233","DOI":"10.1016\/s1364-6613(99)01327-3","article-title":"Is imitation learning the route to humanoid robots?","volume":"3","author":"Schaal","year":"1999","journal-title":"Trends Cognitive Sciences"},{"key":"B33","doi-asserted-by":"publisher","first-page":"1110086","DOI":"10.3389\/fpsyg.2023.1110086","article-title":"Effects of regulated learning scaffolding on regulation strategies and academic performance: a meta-analysis","volume":"14","author":"Shao","year":"2023","journal-title":"Front. Psychol."},{"key":"B34","first-page":"13139","article-title":"Language-conditioned imitation learning for robot manipulation tasks","volume":"33","author":"Stepputtis","year":"2020","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"B35","doi-asserted-by":"crossref","first-page":"18","DOI":"10.1109\/ICARA64554.2025.10977578","article-title":"Lav-act: language-augmented visual action chunking with transformers for bimanual robotic manipulation","volume-title":"2025 11th International Conference on Automation, Robotics, and Applications (ICARA)","author":"Tripathi","year":"2025"},{"key":"B36","doi-asserted-by":"publisher","first-page":"271","DOI":"10.1007\/s10648-010-9127-6","article-title":"Scaffolding in teacher\u2013student interaction: a decade of research","volume":"22","author":"Van de Pol","year":"2010","journal-title":"Educ. Psychology Review"},{"key":"B37","first-page":"5998","article-title":"Attention is all you need","volume":"30","author":"Vaswani","year":"2017","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"B38","article-title":"Roboeval: where robotic manipulation meets structured and scalable evaluation","author":"Wang","year":"2025"},{"key":"B39","first-page":"522","article-title":"An auxiliary task boosted multi-task learning method for service account retrieval with limited human annotation","volume-title":"Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing: Industry Track","author":"Yao","year":"2023"},{"key":"B40","article-title":"Licorice: label-efficient concept-based interpretable reinforcement learning","author":"Ye","year":"2025","journal-title":"ICLR"},{"key":"B41","doi-asserted-by":"publisher","first-page":"7173","DOI":"10.1109\/TCYB.2024.3395626","article-title":"A survey of imitation learning: Algorithms, recent developments, and challenges","volume":"54","author":"Zare","year":"2024","journal-title":"IEEE Trans. Cybern."},{"key":"B42","doi-asserted-by":"publisher","first-page":"016","DOI":"10.15607\/RSS.2023.XIX.016","article-title":"Learning fine-grained bimanual manipulation with low-cost hardware","author":"Zhao","year":"2023","journal-title":"Robotics Sci. Syst. XIX"}],"container-title":["Frontiers in Robotics and AI"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frobt.2026.1865290\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,8,12]],"date-time":"2026-08-12T04:38:22Z","timestamp":1786509502000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frobt.2026.1865290\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,8,12]]},"references-count":42,"alternative-id":["10.3389\/frobt.2026.1865290"],"URL":"https:\/\/doi.org\/10.3389\/frobt.2026.1865290","relation":{},"ISSN":["2296-9144"],"issn-type":[{"value":"2296-9144","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,8,12]]},"article-number":"1865290"}}