aiqbench analysis
This research addresses the critical challenge of compositional generalization in sequential robot tasks by decomposing the generalization gap into specific instruction-space shifts. The study demonstrates that exhaustive data collection is unnecessary; instead, prioritizing dependency coverage and strategic finetuning can drastically improve out-of-distribution performance while reducing training costs. These findings provide a clear framework for optimizing robotic policy learning without relying on combinatorially expensive demonstration sets.
Diagnosing Compositional Generalization in Sequential Robot Tasks
arXiv:2607.29687v1 Announce Type: new Abstract: Sequential robot manipulation requires policies to execute novel combinations of familiar instruction components. However, collecting demonstrations for all possible instruction tuples is combinatorially expensive, while sparsely covered datasets often fail under out-of-distribution recombination. This paper studies compositional generalization through the lens of instruction-space coverage. We decompose the generalization gap into three sources: \textit{marginal instruction shift}, \textit{instruction-compositional shift}, and \textit{context--action shift}. This decomposition allows us to diagnose when sparse training coverage is sufficient, and what structure the training set must preserve for reliable action prediction. Our results show that exhaustive tuple enumeration is unnecessary: a structured subset, as small as one quarter of the full task space, can recover strong out-of-distribution performance when it covers action-relevant dependencies. We further find that sparse training often fails due to instruction steering rather than missing low-level skills; finetuning only one demonstration per task improves OOD success from \(0.4\%\) to \(54.7\%\). For semantically dependent tasks, effective coverage must capture relational structure rather than only factor diversity. These findings suggest that efficient robot data collection should prioritize dependency coverage in instruction space over exhaustive task expansion. More results are available in the supplementary material. Project website: https://yixiaowang7.github.io/Diagnosing_Compositional_Generalization_Robot_Page/.
Abstract:Sequential robot manipulation requires policies to execute novel combinations of familiar instruction components. However, collecting demonstrations for all possible instruction tuples is combinatorially expensive, while sparsely covered datasets often fail under out-of-distribution recombination. This paper studies compositional generalization through the lens of instruction-space coverage. We decompose the generalization gap into three sources: \textit{marginal instruction shift}, \textit{instruction-compositional shift}, and \textit{context--action shift}. This decomposition allows us to diagnose when sparse training coverage is sufficient, and what structure the training set must preserve for reliable action prediction. Our results show that exhaustive tuple enumeration is unnecessary: a structured subset, as small as one quarter of the full task space, can recover strong out-of-distribution performance when it covers action-relevant dependencies. We further find that sparse training often fails due to instruction steering rather than missing low-level skills; finetuning only one demonstration per task improves OOD success from \(0.4\%\) to \(54.7\%\). For semantically dependent tasks, effective coverage must capture relational structure rather than only factor diversity. These findings suggest that efficient robot data collection should prioritize dependency coverage in instruction space over exhaustive task expansion. More results are available in the supplementary material. Project website: this https URL.
Robotics (cs.RO)
arXiv:2607.29687 [cs.RO]
(or arXiv:2607.29687v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2607.29687 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 286 words
Related coverage
ATOM-Bench: A Real-World Benchmark for Atomic Skills and Compositional Generalization in Manipulation Policies
arXiv · Jun 16, 2026
RoboFoundry: System-as-Policy Evolution for Self-Learning Embodied Agents
arXiv · Sep 29, 2026
InstructMove: A Text-Indispensable Benchmark for Instruction-Following Manipulation
arXiv · Aug 25, 2026
Paired Exact-Reset Evaluation of a Prediction-Derived Medium-to-Full World-Model Cascade
arXiv · Aug 19, 2026