Exploiting Embodied Simulation to Detect Novel Object Classes Through Interaction

TL;DR

This paper introduces a method for naive agents to detect novel objects by analyzing interaction outcomes using learned embeddings from reinforcement learning policies, enabling the identification of new object classes.

Contribution

The paper presents a novel approach that leverages embodied simulation and embedding analysis to detect object novelty without prior explicit knowledge of new classes.

Findings

Effective detection of novel objects using learned embeddings

Method works across different datasets and policies

Provides insights into environmental information needed for novelty detection

Abstract

In this paper we present a novel method for a naive agent to detect novel objects it encounters in an interaction. We train a reinforcement learning policy on a stacking task given a known object type, and then observe the results of the agent attempting to stack various other objects based on the same trained policy. By extracting embedding vectors from a convolutional neural net trained over the results of the aforementioned stacking play, we can determine the similarity of a given object to known object types, and determine if the given object is likely dissimilar enough to the known types to be considered a novel class of object. We present the results of this method on two datasets gathered using two different policies and demonstrate what information the agent needs to extract from its environment to make these novelty judgments.