Document Type

Restricted

Advisor

Gary Parker

Publication Date

2026

Comments

This paper is restricted to the Connecticut College campus until December 8, 2026.

Abstract

Evolutionary computation has proven to be a viable solution for various problems in the field of robotics. However, evolutionary computation is dependent on testing numerous solutions over many generations. Over time, this will prove to be damaging to robot hardware. This issue is compounded with more difficult and practical solutions. Simulations have been used in place of real robots for this reason, but simulations can be inaccurate and solutions may not be perfectly transferable to a real robot when deployed in practice. Punctuated anytime learning (PAL) presents a solution to this problem by bridging the gap between simulation and the real robot by doing some learning in a simulation. PAL periodically tests those solutions on the real robot which informs the continued learning process in the simulation. In this work, we implement a PAL system on a real robot, doing so for the first time with no human interaction during the learning process.

Download

Share

COinS
 

The views expressed in this paper are solely those of the author.