Reconciling open-ended self-adaptation with trustworthiness by design — enabling robots to adapt their behaviour to drastic and unpredicted changes while maintaining safety, robustness and performance
Project Duration: 36 months · Consortium: 10 Partners from 8 Countries
The following figures are extracted from the RoboSAPIENS project proposal, illustrating the core architecture, methodology, and work package structure.
Comparison across five dimensions: Trustworthiness & Safety Assurance, Levels of Adaptivity, Correctness of Techniques, Deep Learning, and Active Uncertainty Reduction. Each objective pushes limits in two or more dimensions simultaneously.
Left: traditional unsafe MAPE-K self-adaptation. Right: RoboSAPIENS adds a Validate Safety & Performance function and a MAPE-K Trustworthiness Checker for open-ended safe and trustworthy self-adaptation.
Data and adaptation flow between WP1 (Foundations), WP2 (Safe Self-Adaptation), WP3 (Trustworthiness), WP4 (Case Studies), WP5 (Architecture), WP6 (Dissemination), and WP7 (Management).
In RoboSAPIENS, NTNU (Norwegian University of Science and Technology) leads WP4: Industrial Case Studies and contributes to multiple work packages with expertise in digital twins, co-simulation, and maritime systems.
Seven interconnected work packages driving safe and trustworthy robotic self-adaptation
Develop mathematical theories and formal techniques for compositional verification, operational requirements, uncertainty quantification, and prototype verification tools.
Read more →Develop tools for the MAPE-K loop: monitoring, active environmental state analysis, autonomous learning and adaptation, safety and performance validation, and knowledge modelling.
Read more →Develop methods to produce MAPE-K trustworthiness checkers through domain analysis, standard definition, synthesis of verified checkers, and integration with the adaptation loop.
Read more →Evaluate adaptation technologies across four domains: robot disassembling, robot navigation, ship motion prediction, and dynamic risk modelling in Industry 4.0.
Read more →Develop the RoboSAPIENS platform integrating results from WP2 and WP3 into a reference embedded architecture for robotic adaptivity, with workflow and elaborated examples.
Read more →Raise awareness, disseminate results among stakeholders, liaise with Digital Innovation Hubs, and lay out the roadmap for exploiting project results through the INTO-CPS Association.
Read more →Overall project management including quality assurance, risk management, innovation monitoring, requirements management, and progress and cost reporting.
Read more →Four industry-scale use cases demonstrating safe robotic self-adaptation across diverse domains
Self-adapting force-based manipulation primitives for human-robot collaboration in laptop disassembly remanufacturing, adapting to parts in different conditions using expert demonstrations.
Fleet of TIAGo robots on the shop floor with dynamic self-adaptation for navigation, task scheduling, and safety when robots enter/leave the fleet or floor plans change.
Transfer learning for ship motion prediction using the Gunnerus research vessel, combining dynamic models with limited data to adapt to structural changes like hull prolongation.
Dynamic human-robot safety model using sensor data and machine learning to automatically assess and adapt risk in collaborative assembly tasks in Industry 4.0.
RoboSAPIENS pushes the state-of-the-art across five interconnected dimensions
Incorporate functional safety measures into the MAPE-K loop, ensuring system-level safety properties can be verified and monitored during self-adaptation.
Tackle both structural and environmental changes including humans, with safety observation and validation split across the robot and its runtime Digital Twin.
Formalise the layered robotic architecture and MAPE-K architecture with unified formal semantics handling data, reaction, time, and uncertainty.
Develop active learning and active perception approaches, addressing catastrophic forgetting and distribution shifts in a holistic DL-based self-adaptation pipeline.
Develop novel uncertainty quantification metrics for DL models in robotic systems, with mechanisms to actively reduce uncertainties during operation.
For questions about the RoboSAPIENS project, reach out to our coordination team
Aarhus University (AU)
Department of Electrical and Computer Engineering
Denmark
Prof. Houxiang Zhang
NTNU — Norwegian University of Science and Technology
Department of Ocean Operations and Civil Engineering
Larsgårdsveien 2, 6009 Ålesund, Norway
Email: hozh@ntnu.no
Horizon Europe — HORIZON-CL4-2023-DIGITAL-EMERGING-01-01
Novel paradigms and approaches, towards AI-driven autonomous robots
A strong consortium of 10 partners from 8 countries across academia, research institutes, and industry