Challenge Project
R-ADA
Building a rational, automated robot design agent
A rational automated design agent for robotics, combining language models, simulation, probabilistic programming, and Bayesian inference.
Overview
R-ADA is a Rational Automated Design Agent for robotics. The project aims to automate robot design by treating morphology and control as a sequential design process, closer to how human engineers reason through candidate designs than to brute-force search over fixed parameter spaces.
R-ADA combines language models that can work with CAD/CAM environments, simulation-based evaluation, probabilistic methods for uncertainty, and Bayesian updating to reason about the gap between simulation and real-world performance.
Approach
The system is designed to explore large, discontinuous design spaces while still considering practical constraints such as manufacturability, maintenance, and real-world robustness.