Publicação
BT-Enabled cognitive architecture: physical world sensing-processing-acting cycle/subsystem
| Resumo: | As the Artificial Intelligence (AI) domain expands, there is an increasing need for advancements that go beyond the conventional data-driven approaches and embrace autonomous learning, drawing inspiration from the learning abilities and deep understanding of the world exhibited by the biological brain. Cognitive Architectures (CAs) have emerged as a crucial area of research, striving to model human-like cognition and reasoning. Nevertheless, the complex nature of cognitive systems poses challenges in integrating their diverse modules, impacting their explainability and expandability. This dissertation focuses on developing a subsystem within a CA that interfaces an agent with the physical world, employing Behavior Trees (BTs) as the underlying structuring mechanism. The proposed solution revolves around a sensing-processing-acting cycle, where the constituent modules establish connections with the diverse memory structures of the agent, empowering it to perceive, learn, decide, and act accordingly. The generic design specifications of this solution are geared towards a use case that facilitates the contextualized demonstration of this solution within a practical scenario. The selected use case consists of an agent navigating an unfamiliar environment, actively perceiving and recognizing relevant points of interest, referred to as references, as well as their interconnections. The agent selects logical routes, leveraging its own knowledge, and seeks assistance when faced with unknown paths. The efficiency of this solution is demonstrated through the implementation of the subsystem on a low end embedded system. Supporting the subsystem is a custom BT engine optimized for embedded system execution, complemented by a monitoring tool that enables real-time observation of BT execution. Validation of this work is achieved through simulations conducted on a real prototype deployed on an embedded platform, operating in a controlled environment to allow the generation of prompt and well founded conclusions. The results demonstrate the potential of the proposed approach in bridging the gap between current AI systems and the remarkable learning abilities observed in biological systems. Furthermore, they affirm the scalability of the developed CA, both within the use case under consideration and in other diverse applications. |
|---|---|
| Autores principais: | Silva, João Manuel Gonçalves |
| Assunto: | Cognitive architectures Behavior trees Embedded systems Arquiteturas cognitivas Árvores de comportamento Sistemas embebidos |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | As the Artificial Intelligence (AI) domain expands, there is an increasing need for advancements that go beyond the conventional data-driven approaches and embrace autonomous learning, drawing inspiration from the learning abilities and deep understanding of the world exhibited by the biological brain. Cognitive Architectures (CAs) have emerged as a crucial area of research, striving to model human-like cognition and reasoning. Nevertheless, the complex nature of cognitive systems poses challenges in integrating their diverse modules, impacting their explainability and expandability. This dissertation focuses on developing a subsystem within a CA that interfaces an agent with the physical world, employing Behavior Trees (BTs) as the underlying structuring mechanism. The proposed solution revolves around a sensing-processing-acting cycle, where the constituent modules establish connections with the diverse memory structures of the agent, empowering it to perceive, learn, decide, and act accordingly. The generic design specifications of this solution are geared towards a use case that facilitates the contextualized demonstration of this solution within a practical scenario. The selected use case consists of an agent navigating an unfamiliar environment, actively perceiving and recognizing relevant points of interest, referred to as references, as well as their interconnections. The agent selects logical routes, leveraging its own knowledge, and seeks assistance when faced with unknown paths. The efficiency of this solution is demonstrated through the implementation of the subsystem on a low end embedded system. Supporting the subsystem is a custom BT engine optimized for embedded system execution, complemented by a monitoring tool that enables real-time observation of BT execution. Validation of this work is achieved through simulations conducted on a real prototype deployed on an embedded platform, operating in a controlled environment to allow the generation of prompt and well founded conclusions. The results demonstrate the potential of the proposed approach in bridging the gap between current AI systems and the remarkable learning abilities observed in biological systems. Furthermore, they affirm the scalability of the developed CA, both within the use case under consideration and in other diverse applications. |
|---|