Fast, robust, and versatile event detection through HMM belief state gradient measures

Abstract

Event detection is a critical feature in data-driven systems as it assists with the identification of nominal and anomalous behavior. Event detection is increasingly relevant in robotics as robots operate with greater autonomy in increasingly unstructured environments. In this work, we present an accurate, robust, fast, and versatile measure for skill and anomaly identification. A theoretical proof establishes the link between the derivative of the log-likelihood of the HMM filtered belief state and the latest emission probabilities. The key insight is the inverse relationship in which gradient analysis is used for skill and anomaly identification. Our measure showed better performance across all metrics than all but one related stateof-the-art works. The result is broadly applicable to domains that use HMMs for event detection. Supplemental information, code, data, and videos can be found at [1].

Nonparametric Bayesian Method for Robot Anomaly Monitoring

“…Anomaly identification sensitivity is intimately connected to threshold setting. We compare the results of our execution varying threshold with a fixed threshold presented in [35]. The comparison is seen in Fig.…”

An Online Robot Collision Detection and Identification Scheme by Supervised Learning and Bayesian Decision Theory

“…As alternatives, probabilistic series-models, such as the Hidden Markov Model (HMM) [32] and the Gaussian Mixture Model (GMM) [33] are also applied to CDI by exploiting the dependence properties of time series. In [34], an HMM is developed to detect exceptional events in an object-alignment robot task, where the measured torques and their derivatives are used.…”

A Latent State-Based Multimodal Execution Monitor with Anomaly Detection and Classification for Robot Introspection

“…Anomaly identification sensitivity is intimately connected to threshold setting. We compare the results of our execution varying threshold with a fixed threshold presented in [2]. The comparison is seen in Figure 9.…”

“…For the anomaly detection, we achieved it by comparing the cumulative log-likelihood deviation to a given threshold from nominal executions [1,2,9]. An improved anomaly detector will be proposed in this paper, which can more effectively and robustly detect the anomalies.…”

“…Left column: results of our hidden-state-based anomaly detector. Right column: results of the gradient-based anomaly detector[2]. The ground truth (anomaly_t_by_human) is represented by green dashed lines and the detected anomalies are represented by yellow solid lines.…”