From Object-Action to Property-Action: Learning Causally Dominant Properties through Cumulative Explorative Interactions
Authors: Mohan V., Bhat A. A., Morasso P., & Sandini G.
Journal: Biologically Inspired Cognitive Architectures
Tags: causal, robotics, cognitive, affordance
Link: URL
Abstract:
We explore how robots can discover which object properties drive task outcomes by cumulatively interacting with objects varying along multiple dimensions. Through a dynamic neural field model, robots infer causally dominant properties from exploration data without supervision.
Methodology:
A dynamic field theory-based architecture tracks instantaneous perceptual activations and integrates them across trials. Simulations and real‐robot trials show that the system correctly identifies weight or shape as the critical dimension for floating tasks.