The paper "Generative Expressive Robot Behaviors using Large Language Models" by Karthik Mahadevan et al. proposes a novel approach called Generative Expressive Motion (GenEM) to generate expressive robot behaviors. GenEM leverages large language models (LLMs) to translate human language instructions into parametrized control code, enabling robots to perform a wide range of expressive behaviors adaptively and compositely. The approach uses few-shot chain-of-thought prompting to reason about social norms and generate control code based on the robot's available and learned skills. User studies and simulation experiments demonstrate that GenEM-generated behaviors are perceived as competent and easy to understand by users, with some behaviors even receiving positive feedback in some cases. The paper also highlights the adaptability of GenEM to different types of user feedback and its ability to generate composable behaviors. The authors discuss the limitations of their approach, such as the need for further physical interaction studies and the limitations of current LLMs, and outline future directions for improving expressive behavior generation in robots.The paper "Generative Expressive Robot Behaviors using Large Language Models" by Karthik Mahadevan et al. proposes a novel approach called Generative Expressive Motion (GenEM) to generate expressive robot behaviors. GenEM leverages large language models (LLMs) to translate human language instructions into parametrized control code, enabling robots to perform a wide range of expressive behaviors adaptively and compositely. The approach uses few-shot chain-of-thought prompting to reason about social norms and generate control code based on the robot's available and learned skills. User studies and simulation experiments demonstrate that GenEM-generated behaviors are perceived as competent and easy to understand by users, with some behaviors even receiving positive feedback in some cases. The paper also highlights the adaptability of GenEM to different types of user feedback and its ability to generate composable behaviors. The authors discuss the limitations of their approach, such as the need for further physical interaction studies and the limitations of current LLMs, and outline future directions for improving expressive behavior generation in robots.