Modeling curiosity in a mobile robot for long-term autonomous exploration and monitoring

Abstract

This paper presents a novel approach to modeling curiosity in a mobile robot, which is useful for monitoring and adaptive data collection tasks, especially in the context of long term autonomous missions where preprogrammed missions are likely to have limited utility. We use a realtime topic modeling technique to build a semantic perception model of the environment, using which, we plan a path through the locations in the world with high semantic information content. The life-long learning behavior of the proposed perception model makes it suitable for long-term exploration missions. We validate the approach using simulated exploration experiments using aerial and underwater data, and demonstrate an implementation on the Aqua underwater robot in a variety of scenarios. We find that the proposed exploration paths that are biased towards locations with high topic perplexity, produce better terrain models with high discriminative power. Moreover, we show that the proposed algorithm implemented on Aqua robot is able to do tasks such as coral reef inspection, diver following, and sea floor exploration, without any prior training or preparation.

Online informative path planning for active classification using UAVs

“…Non-adaptive approaches explore an environment using sequence of pre-determined actions. Adaptive approaches [13][14][15] exploit new measurements based on specific interests. Comparably to Low et al [16], we use a finite look-ahead, allowing us to propagate map changes and plan adaptively.…”

Adaptive continuous‐space informative path planning for online environmental monitoring

“…An entirely different approach is to formulate reactive behaviors based on curiosity. Girdhar and Dudek have implemented a curiosity‐based visual exploration scheme on an autonomous underwater vehicle (AUV) to collect imagery in shallow waters. Most recently, Lim et al .…”

Anomaly detection in unstructured environments using Bayesian nonparametric scene modeling

“…ROST does this by generalizing the idea of a document to a spatiotemporal cell, and computing the topic label for the words in a cell in the context of its neighboring cells. In [14] we used ROST to identify interesting observations in a robot's view, and then used it to plan an adaptive path, which were shown to have higher information content than simple space filling paths.…”