The paper proposes a language-based programming assistant for industrial robots. The assistant provides interactive dialogues about skills, example use-cases and expected robot behavior. The authors present three alternative model families, trained by three different domain-specific fine-tuning approaches. The performance of the variants is evaluated by comparing their BERTScore performance as well as a user survey with industry experts. The research highlights the potential of large-scale pretrained foundation models such as LLaMA for use in various application domains.
Publication date: 21 Dec 2023
Project Page: https://arxiv.org/abs/2312.13905v1
Paper: https://arxiv.org/pdf/2312.13905
