To support us in our daily lives, future humanoid assistant robots need both a certain degree of autonomy to learn and explore, but must also adhere to certain constraints. Constrained planning is a problem especially for otherwise promising approaches based on neural networks - while symbolic approaches are very good at adhering to constraints, but have trouble with complex, new environments. New developments in the area of neuro-symbolic AI would revolutionize constrained task planning. On this year's Symposium on Access Control Models and Technologies, we have presented our BlueSky publication “How to Raise a Robot - A Case for Neuro-Symbolic AI in Constrained Task Planning for Humanoid Assistive Robots” on that topic. The publication is freely available as open access under CC-BY.

We presented our new publication “Logical Clocks and Monotonicity for Byzantine-Tolerant Replicated Data Types” at this year's Workshop on Principles and Practice of Consistency for Distributed Data. In the paper, we decompose Matrix and other autonomous decentralized systems into their order-theoretic foundations. We show that the decomposition allows to analyze Byzantine fault tolerance on the basis of the order-theoretic property of monotonicity. The article is freely available as open access under CC-BY

Chaincode

At the Bitcoin Research Day in Fall 2024 in New York, Prof. Dr. Hannes Hartenstein and Matthias Grundmann will give a talk about our research. The talk is titled "From Monitoring Bitcoin to Verifying Lightning: Modeling Aspects of the Bitcoin Universe". Find more information at brd.chaincode.com.

For our SACMAT poster, there is now a follow-up extended version on arXiv under the subtitle “A Case for Neuro-Symbolic AI in Constrained Task Planning for Humanoid Assistive Robots”. We report on new experiments with the unification of robot task planning and access control via Large Language Models, and show how different levels of neuro-symbolic integration can further strengthen this unification.