We invite submissions to the IROS 2026 Workshop on Hybrid Architectures for Embodied Autonomy: Bridging Learning, Planning, and Control.
Submissions will be selected based on technical merit, innovation, relevance, clarity of presentation and potential impact.
All submissions must be made to OpenReview.
Submission Format
Extended abstract: maximum 4 pages of main content, excluding references
Papers must be submitted in PDF format and must strictly adhere to the official IEEE RAS double-column format. You can find the templates on the official IEEE RAS template page or via the IROS 2026 call for papers guidelines.
Optional: 2-3 minutes video presentation
The workshop is non-archival, and we welcome original work, preliminary results, position papers, benchmarks, negative results, and recently published work of clear relevance to the workshop. Accepted contributions will be presented as posters, with selected submissions invited for spotlight talks.
Important Dates
Submission opens: June 14, 2026
Submission deadline: August 14, 2026, 23:59 AoE
Acceptance Notification: September 1, 2026
Final/camera-ready submission deadline: September 11, 2026, 23:59 AoE
Workshop date: September 27 or October 1, 2026, IROS 2026, Pittsburgh, PA
Representative Topics
We welcome contributions that study how classical structure can support learning-based autonomy, how learning can improve planning and control pipelines, and how these components can be composed into reliable autonomy stacks. Topics of interest include, but are not limited to:
hybrid autonomy architectures and interfaces between learned and model-based components;
classical structure as scaffolding for robot learning, including models, constraints, dynamics, geometry, optimization, verification, and certificates;
learning-based components that make planning and control pipelines more flexible, adaptive, or robust;
deployment-oriented evaluation protocols that go beyond task success to assess robustness, safety, latency, data efficiency, sim-to-real transfer, and failure recovery.
We are especially interested in work that moves beyond isolated algorithmic improvements and helps clarify the architectural choices, interface assumptions, and real-world implications of integrating learning with classical methods for planning and control.