Accelerated co-design of robots through morphological pretraining
Luke Strgar, Sam Kriegman · 2025
Read Accelerated co-design of robots through morphological pretraining by Luke Strgar, Sam Kriegman (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-10862.txt.
Description
Luke Strgar and Sam Kriegman speed up the joint design of robot bodies and robot brains. The usual route trains a fresh controller for every body shape, which eats enormous data. Their shortcut first trains one universal controller that can steer many shapes, using a simulator smooth enough for direct gradient tuning.
With that pretrained controller in hand, a designer can add, remove, or swap body parts and instantly see whether the change helps, a process they call zero-shot evolution. Evolving bodies and controller together, by contrast, collapsed the population into a few similar shapes that were easy to steer, starving the controller of variety. Zero-shot evolution quickly produced diverse, high-performing designs. Fine-tuning the shared controller as evolution ran preserved the variety and pushed performance higher still.
Catalog record
| CREATOR | Luke Strgar, Sam Kriegman |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-10862