Abstract
ABSTRACT/ SUMMARY The ENGINES (ENGineering INtElligent Systems around intelligent agent technologies) project was conceived to address a critical fragmentation in modern Intelligent Systems Engineering (ISE). While engineers today have access to an overwhelming array of Artificial Intelligence (AI) techniques, these are predominantly utilized in isolation or through rigid, ad-hoc integrations tailored to highly specific applications. This lack of a cohesive methodology hinders reusability, explainability, and conceptual integrity. To overcome this, the overarching goal of the project was to define a comprehensive conceptual and technical framework (providing models, guidelines, and tools) that synergistically integrates diverse AI techniques using intelligent agent technologies as a common grounding. The project adopted the Belief-Desire-Intention (BDI) intelligent agent architecture as its foundational standard. BDI was selected due to its proven efficacy in engineering autonomous entities capable of rational, explainable, and goal-directed reasoning in dynamic environments. The original objective was to augment this robust symbolic architecture with selected AI capabilities, specifically focusing on coordination and agreement, conversation and natural language processing (NLP), spatial modeling, and semantic knowledge representation. Crucially, during its execution, the project successfully adapted to the disruptive paradigm shift brought about by the advent of Large Language Models (LLMs) and Generative AI. Recognizing the revolutionary impact of "Agentic AI," the consortium strategically evolved the project's scope to pioneer a highly effective hybrid approach. Instead of relying solely on traditional models, the project rigorously integrated Machine Learning (ML) and LLM technologies into the BDI framework. This enabled the creation of hybrid systems that combine the extreme flexibility and conversational fluency of neural networks with the deterministic reliability, rule enforcement, and explainability of symbolic BDI reasoning. RESULTS ACHIEVED/ OUTPUTS In alignment with its expected outcomes, the project delivered tangible results across four main pillars: 1. Extended architectures and models: the consortium developed novel architectures that augment classical BDI agents with LLM capabilities and spatial awareness, enabling deployment in both Virtual Reality environments (e.g., VEsNA, ChatBDI) and resource-constrained IoT edge devices (e.g., DEMOCLE). 2. Guidelines and theoretical foundations: the consortium produced foundational research, including comprehensive surveys mapping ML-BDI integration, methodologies for enforcing cultural rules within LLMs, and visionary roadmaps for decentralized multi-agent AI. 3. Toolchains for engineering: the consortium realized practical developer tools, including advanced wireless communication interfaces for edge networks (Hermes) and enhanced network simulation environments (LWN Simulator). 4. Proofs of concept: the consortium validated the theoretical frameworks through real-world applications, such as a robust indoor positioning and assistance system utilizing LLM natural language interfaces, and autonomous avatars capable of situated communication. PREDICTIVE ANALYSIS/ FINAL COMMENTS The framework and technologies developed during the ENGINES project serve as a foundational springboard for several highly promising future scenarios. Because the consortium successfully pivoted to integrate Large Language Models (LLMs) with Belief-Desire-Intention (BDI) architectures, the resulting hybrid systems are perfectly positioned to address the current market and scientific demand for reliable AI agents. Specifically, we forecast the following concrete developments stemming from the project's outcomes: 1. The theoretical roadmap and architectural models (particularly those systematizing ML/BDI integration and rule-enforcement) provide a highly competitive baseline for upcoming European funding calls (e.g., Horizon Europe). Future proposals will likely focus on scaling these decentralized, multi-agent AI frameworks for critical infrastructure and trustworthy AI initiatives. 2. The practical toolchains developed (such as the Hermes mesh network interface, the lightweight DEMOCLE logic engine, and the enhanced LWN Simulator) have strong commercial viability. We anticipate future technological transfer activities applying these IoT and edge-computing solutions to smart city logistics, industrial safety monitoring, and decentralized sensor networks. 3. The ChatBDI and VEsNA frameworks open direct pathways for developing enterprise-grade Digital Twins and intelligent virtual assistants. Future projects will likely expand on these prototypes to create fully autonomous, context-aware avatars for use in healthcare, education, and virtual reality training environments. Looking back at the project's lifecycle, the most significant lesson learned is the absolute necessity of scientific agility. In fast-paced technological sectors like AI, rigid adherence to a multi-year project plan can lead to obsolescence. The ENGINES consortium turned a potential threat (the disruptive advent of Generative AI, which threatened to overshadow classical BDI approaches) into our greatest opportunity. By proactively embracing LLMs, we successfully delivered a state-of-the-art hybrid architecture rather than a legacy software project. The primary strength of the project was this adaptability, coupled with the highly complementary expertise of the operational units (ranging from theoretical formalization to edge-computing engineering). The integration of generative models into deterministic frameworks remains computationally demanding. A continuing challenge (and an area for necessary future change) is the heavy resource footprint of LLMs, which still complicates their seamless deployment on highly constrained edge devices without relying on cloud APIs. Furthermore, establishing standardized evaluation metrics for these hybrid neuro-symbolic systems remains an open challenge for the scientific community. Ultimately, the ENGINES project demonstrates that the future of AI does not lie in abandoning symbolic logic for pure machine learning, but rather in their synthesis. The resulting architectures offer a sustainable, explainable, and highly capable paradigm that will define the next generation of autonomous intelligent systems.
Dettagli del progetto
Responsabile scientifico: Andrea Omicini
Strutture Unibo coinvolte:
Dipartimento di Informatica - Scienza e Ingegneria
Coordinatore:
ALMA MATER STUDIORUM - Università di Bologna(Italy)
Contributo totale di progetto: Euro (EUR) 213.185,00
Contributo totale Unibo: Euro (EUR) 56.520,00
Durata del progetto in mesi: 29
Data di inizio
28/09/2023
Data di fine:
28/02/2026
