Pretrained Embeddings as a Behavior Specification Mechanism

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Kapoor, Parv, Hammer, Abigail, Kapoor, Ashish, Leung, Karen, Kang, Eunsuk
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866929744182771712
author Kapoor, Parv
Hammer, Abigail
Kapoor, Ashish
Leung, Karen
Kang, Eunsuk
author_facet Kapoor, Parv
Hammer, Abigail
Kapoor, Ashish
Leung, Karen
Kang, Eunsuk
contents We propose an approach to formally specifying the behavioral properties of systems that rely on a perception model for interactions with the physical world. The key idea is to introduce embeddings -- mathematical representations of a real-world concept -- as a first-class construct in a specification language, where properties are expressed in terms of distances between a pair of ideal and observed embeddings. To realize this approach, we propose a new type of temporal logic called Embedding Temporal Logic (ETL), and describe how it can be used to express a wider range of properties about AI-enabled systems than previously possible. We demonstrate the applicability of ETL through a preliminary evaluation involving planning tasks in robots that are driven by foundation models; the results are promising, showing that embedding-based specifications can be used to steer a system towards desirable behaviors.
format Preprint
id arxiv_https___arxiv_org_abs_2503_02012
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Pretrained Embeddings as a Behavior Specification Mechanism
Kapoor, Parv
Hammer, Abigail
Kapoor, Ashish
Leung, Karen
Kang, Eunsuk
Artificial Intelligence
Robotics
Software Engineering
We propose an approach to formally specifying the behavioral properties of systems that rely on a perception model for interactions with the physical world. The key idea is to introduce embeddings -- mathematical representations of a real-world concept -- as a first-class construct in a specification language, where properties are expressed in terms of distances between a pair of ideal and observed embeddings. To realize this approach, we propose a new type of temporal logic called Embedding Temporal Logic (ETL), and describe how it can be used to express a wider range of properties about AI-enabled systems than previously possible. We demonstrate the applicability of ETL through a preliminary evaluation involving planning tasks in robots that are driven by foundation models; the results are promising, showing that embedding-based specifications can be used to steer a system towards desirable behaviors.
title Pretrained Embeddings as a Behavior Specification Mechanism
topic Artificial Intelligence
Robotics
Software Engineering
url https://arxiv.org/abs/2503.02012