EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

ConceptACT: episode-level concepts for sample-efficient robotic imitation learning

Jakob Karalus, Friedhelm Schwenker

Frontiers in Robotics and AI · 12.08.2026 · 13

DOI: 10.3389/frobt.2026.1865290 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Frontiers in Robotics and AI, online erschienen am 12.08.2026. Verfasst von Jakob Karalus und 1 weiteren Autor:innen. Beteiligte Einrichtung: Institute of Artificial Intelligence, Ulm University, Ulm, Germany. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

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.

Schlagwörter

Stichwörter der Autor:innen

action chunkingattention mechanismsconcept learninghuman-robot interactionimitation learningrobotic manipulationsample efficiencytransformer networks

Autor:innen und Institutionen

  • Institute of Artificial Intelligence, Ulm University, Ulm, Germany.
  • Institute of Neuroinformatics, Ulm University, Ulm, Germany.

Zitieren

APA

Karalus, J., & Schwenker, F. (2026). ConceptACT: episode-level concepts for sample-efficient robotic imitation learning. Frontiers in Robotics and AI, 13. https://doi.org/10.3389/frobt.2026.1865290

Vancouver

Karalus J, Schwenker F. ConceptACT: episode-level concepts for sample-efficient robotic imitation learning. Frontiers in Robotics and AI. 2026;13. doi:10.3389/frobt.2026.1865290

BibTeX
@article{Karalus2026,
  author = {Karalus, Jakob and Schwenker, Friedhelm},
  title = {ConceptACT: episode-level concepts for sample-efficient robotic imitation learning},
  journal = {Frontiers in Robotics and AI},
  year = {2026},
  volume = {13},
  doi = {10.3389/frobt.2026.1865290}
}
Kostenlos starten →