Learning Low-Level Causal Relations using a Simulated Robotic Arm

TL;DR

This paper explores learning low-level causal relations in a simulated robotic arm through forward and inverse models, analyzing feature attribution for better understanding and explainability of causal effects in sensorimotor tasks.

Contribution

It introduces a method for learning and analyzing low-level causal relations in robotic systems using simulated data and feature attribution techniques.

Findings

Feature attribution reveals low-level causal effects of individual state features.

Analysis supports dimensionality reduction and improved explainability.

Demonstrates causal learning in simulated robotic sensorimotor tasks.

Abstract

Causal learning allows humans to predict the effect of their actions on the known environment and use this knowledge to plan the execution of more complex actions. Such knowledge also captures the behaviour of the environment and can be used for its analysis and the reasoning behind the behaviour. This type of knowledge is also crucial in the design of intelligent robotic systems with common sense. In this paper, we study causal relations by learning the forward and inverse models based on data generated by a simulated robotic arm involved in two sensorimotor tasks. As a next step, we investigate feature attribution methods for the analysis of the forward model, which reveals the low-level causal effects corresponding to individual features of the state vector related to both the arm joints and the environment features. This type of analysis provides solid ground for dimensionality reduction of the state representations, as well as for the aggregation of knowledge towards the explainability of causal effects at higher levels.