Conjecture—Dual Process Theory as a path forward for robust real-world robot learning

Assess whether adopting architectures inspired by Dual Process Theory can address overconfidence, out-of-distribution risks, and the inability of current learned models to know when they do not know in real-world robotics.

Background

Robots often deploy overconfident models that fail under distribution shifts and lack mechanisms to recognize and remediate uncertainty.

The authors conjecture that Dual Process Theory can guide architectural solutions to these challenges, combining fast intuition with deliberate reasoning and metacognition.