Capability-Driven Skill Generation with LLMs: A RAG-Based Approach for Reusing Existing Libraries and Interfaces

Abstract: Modern automation systems increasingly rely on modular architectures, with capabilities and skills as one solution approach. Capabilities define the functions of resources in a machine-readable form and skills provide the concrete implementations that realize those capabilities. However, the development of a skill implementation conforming to a corresponding capability remains a time-consuming and challenging task. In this paper, we present a method that treats capabilities as contracts for skill implementations and leverages LLMs to generate executable code based on natural language user input. A key feature of our approach is the integration of existing software libraries and interface technologies, enabling the generation of skill implementations across different target languages. We introduce a framework that allows users to incorporate their own libraries and resource interfaces into the code generation process through a retrieval-augmented generation architecture. The proposed method is evaluated using an autonomous mobile robot controlled via Python and ROS 2, demonstrating the feasibility and flexibility of the approach.

Capability-Driven Skill Generation with LLMs: A RAG-Based Approach for Reusing Existing Libraries and Interfaces

The paper by Vieira da Silva et al. proposes a novel method for generating executable robotic skills from abstract capability descriptions using LLMs and a Retrieval-Augmented Generation (RAG) approach. This methodology presents a compelling solution to automation and integration challenges in industrial systems by leveraging the synergy between semantic capability modeling and contextual retrieval of resource interface documentation.

Overview

Modern automation systems necessitate modularity and flexibility due to varying tasks and configurations, compelling the rise of capabilities and skills as pivotal concepts. Capabilities offer abstract descriptions of resource functions, while skills deliver the concrete implementations for execution. However, skill development remains a laborious task, requiring manual code writing and deep technical knowledge of target resources.

The authors address this challenge by employing LLMs to automatically generate skill implementations from ontological capability descriptions. A distinctive feature of this methodology is the RAG-based approach that integrates existing libraries and interfaces to facilitate code synthesis across diverse target languages and frameworks.

Methodology

The proposed method, termed LLMCap2Skill, involves three key steps: user input, API documentation generation, and skill generation using RAG.

1. User Input: The method requires several inputs, including the capability ontology, a skill specification, and target programming details. This provides a structural basis and behavioral context for the skill generation.
2. API Documentation Generation: To enable resource control, the method automatically generates a structured API from the resource's available interfaces. This involves the introspection of the resource to identify relevant topics, services, and actions—such as publishing to or subscribing from specific interfaces.
3. Skill Generation: Utilizing the RAG architecture, the method conducts a similarity search to identify relevant resource interfaces. These interfaces, alongside the capability and user-specified behavior, are used to construct a detailed prompt for the LLM. The LLM then generates a skill implementation conforming to the defined behavior.

Evaluation

The approach was evaluated using a simulated autonomous mobile robot, demonstrating its ability to generate syntactically correct and largely functionally accurate skills. The evaluation considered both structural correctness and functional execution, with skills tested in Gazebo simulations. The method shows effectiveness in reducing manual effort, although minor adjustments for specific cases, such as embedding robot-specific identifiers and refining control logic, were necessary.

Implications and Future Prospects

This research underscores the potential of LLMs to automate robotic skill generation significantly, reducing engineering time while promoting scalability in dynamic environments. By embedding capability models into the skill development process, the approach bridges the gap between abstract specification and executable reality.

Future developments could expand the framework to accommodate diverse automation environments, incorporate richer context for interface retrieval, and integrate formal verification for safety-critical applications. Emphasizing simulation feedback and user interaction could further refine generated skills for real-world deployment.

In conclusion, Vieira da Silva et al. offer a robust framework for leveraging LLMs in automation, opening avenues for further exploration in machine-enabled skill development, thereby addressing pressing industry needs for efficiency and adaptability.