A carregar...
Publicação
Ferramentas de Inteligência Artificial generativa: Avanços e Desafios no Processamento de Linguagem Natural
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 22.47 MB | Adobe PDF |
Orientador(es)
Resumo(s)
A crescente integração de sistemas baseados em Inteligência Artificial na indústria
tem impulsionado o desenvolvimento de novas formas de interação entre operadores
humanos e sistemas robóticos. Neste contexto, a tradução automática de instruções
em linguagem natural para código executável surge como uma abordagem promissora
para simplificar a programação robótica e aproximar os sistemas industriais dos
princípios da Indústria 5.0, centrados na colaboração humano–máquina.
Esta dissertação investiga o papel das ferramentas de Inteligência Artificial
Generativa na conversão de linguagem natural em código robótico em ambientes
industriais. O trabalho integra três componentes principais. Em primeiro lugar, é
realizada uma revisão sistemática da literatura com o objetivo de analisar o estado
da arte na avaliação de modelos de Inteligência Artificial Generativa aplicados a
tarefas de Processamento de Linguagem Natural e geração automática de código. Em
segundo lugar, é desenvolvida uma plataforma experimental capaz de transformar
instruções em linguagem natural em código robótico executável, num ambiente
de simulação controlado. Por fim, é proposta a ferramenta HARPA, um modelo
conceptual de avaliação multidimensional que combina métricas técnicas, operacionais
e humanas para analisar o desempenho destes sistemas.
Os resultados obtidos indicam que os modelos atuais apresentam elevada capaci
dade para gerar código robótico funcional a partir de instruções em linguagem natural.
No entanto, evidenciam limitações relevantes em dimensões como alinhamento hu
mano, fiabilidade e auditabilidade. Verifica-se ainda que métricas tradicionais de
avaliação são insuficientes para validar os requisitos de segurança, previsibilidade,
transparência e colaboração exigidos pela Indústria 5.0.
A ferramenta HARPA constitui uma abordagem promissora para estruturar a
avaliação de modelos de Inteligência Artificial Generativa aplicados à programação
robótica industrial.
The increasing integration of Artificial Intelligence systems in industrial envi ronments has fostered the development of new forms of interaction between human operators and robotic systems. In this context, the automatic translation of natural language instructions into executable code is a promising approach to simplify robot programming and support the principles of Industry 5.0, which emphasize human–machine collaboration. This dissertation investigates the role of Generative Artificial Intelligence tools in the conversion of natural language instructions into robotic code within industrial contexts. The research integrates three main components. First, a systematic literature review is conducted to analyze the state of the art in the evaluation of Generative Artificial Intelligence models applied to Natural Language Processing and automatic code generation tasks. Second, an experimental platform is developed to transform natural language instructions into executable robotic code within a controlled simulation environment. Third, the HARPA model is proposed as a multidimensional evaluation framework that integrates technical, operational, and human-centered metrics to assess system performance. The experimental results indicate that current models demonstrate strong ca pability in generating functional robotic code from natural language instructions. However, relevant limitations are observed in dimensions such as human alignment, reliability and auditability. Traditional evaluation metrics, frequently focused on textual similarity or syntactic correctness, prove insufficient to validate the requi rements of safety, predictability, transparency, and human–machine collaboration required in industrial environments. The HARPA model offers a structured approach for evaluating Generative Artifi cial Intelligence models applied to robotic programming, aligning technical evaluation with the broader challenges of Industry 5.0.
The increasing integration of Artificial Intelligence systems in industrial envi ronments has fostered the development of new forms of interaction between human operators and robotic systems. In this context, the automatic translation of natural language instructions into executable code is a promising approach to simplify robot programming and support the principles of Industry 5.0, which emphasize human–machine collaboration. This dissertation investigates the role of Generative Artificial Intelligence tools in the conversion of natural language instructions into robotic code within industrial contexts. The research integrates three main components. First, a systematic literature review is conducted to analyze the state of the art in the evaluation of Generative Artificial Intelligence models applied to Natural Language Processing and automatic code generation tasks. Second, an experimental platform is developed to transform natural language instructions into executable robotic code within a controlled simulation environment. Third, the HARPA model is proposed as a multidimensional evaluation framework that integrates technical, operational, and human-centered metrics to assess system performance. The experimental results indicate that current models demonstrate strong ca pability in generating functional robotic code from natural language instructions. However, relevant limitations are observed in dimensions such as human alignment, reliability and auditability. Traditional evaluation metrics, frequently focused on textual similarity or syntactic correctness, prove insufficient to validate the requi rements of safety, predictability, transparency, and human–machine collaboration required in industrial environments. The HARPA model offers a structured approach for evaluating Generative Artifi cial Intelligence models applied to robotic programming, aligning technical evaluation with the broader challenges of Industry 5.0.
Descrição
Palavras-chave
Indústria 5.0 Inteligência Artificial Generativa Processamento de Linguagem Natural Robótica Industrial Avaliação de Modelos HARPA Generative Artificial Intelligence Natural Language Processing Industrial Robotics Industry 5.0 Model Evaluation