Continuous, Automated Validation, and Evaluation of Cognitive Robots in Open-Ended Environments

What technology can enable more agile production, personalized healthcare, efficient logistics, and expand scientific reach into ocean depths? We need cognitive robots that interact with environments, humans, and agents to autonomously acquire skills and perform diverse tasks in open-ended contexts. CAVECORE addresses a critical challenge in AI-enabled robotics: evaluating the quality of robotic agents in open-ended environments, including performance, safety, and reliability. How can we validate robot behavior with trustworthy measurements and automate this task over long periods? CAVECORE aims to streamline the innovation process into goal-oriented, efficient technological progress for organizations developing cognitive robots by training AI robotics researchers in methods for replicable, unbiased validation and evaluation in virtual and real-world environments, allowing humans to calibrate trust towards cognitive robots. CAVECORE enables stakeholders (users, developers, policymakers) to understand the expected and actual quality of cognitive robots, ensuring conformance with emerging ethical, legal, social, and economic requirements of the EU AI Act, and delivering robots capable of continuous self-assessment and improvement. CAVECORE's principle of evaluability-by-design mandates that validation and evaluation concepts be integral to the robot development process, not an afterthought, and that cognitive robots continuously provide evidence of their quality. This principle shapes the training program and expert consortium, which offers interdisciplinary training in cognitive science and robotics, AI metrology, and automated testing and validation.