Can Large Language Models design a Robot?

TL;DR

The paper asks whether ChatGPT-3 can contribute to robot design, a bottleneck beyond language models’ uses in translating instructions and synthesizing code. It presents a human–LLM design process in which the model proposes conceptual options and the human directs choices and implements the design.

• Problem

  The paper asks whether ChatGPT-3 can contribute to robot design, which the authors identify as a robotics bottleneck beyond language models’ uses in translating instructions and synthesizing code.

  Source: p. 1

• Method

  A two-stage human–LLM process develops conceptual robot specifications through dialogue, then has a human translate them into a physical robot.

  Source: pp. 1–2

• Results

  The proposed division has the LLM offer conceptual options and interdisciplinary connections while the human directs choices and handles technical implementation.

  Source: pp. 1–2

• Takeaways & Limitations

  LLM-assisted exploration could augment human expertise and support non-specialists in realizing robotic systems.

  Source: p. 2

  A shared model may steer researchers toward statistically preferred solutions and hinder exploration; the authors also question whether it generates genuinely new creative solutions.

  Source: p. 3