Treffer: Multimodal Prediction-Based Robot Abnormal Movement Identification Under Variable Time-length Experiences.

Robots will eventually make part of our daily lives, helping us at home, taking care of the elderly, and collaborating at work. In such Human-Robot Collaboration (HRC) scenarios, achieving abnormal movement identification can effectively deal with unexpected anomalies such as human collisions, external perturbations, and unexpected changes in the environment. To this end, Long-short Term Memory (LSTMs) based prediction methods are widely proposed for abnormal identification, which typically has an implicit requirement of fixed-length input signals such that the identification thresholds are calculated from the prediction-error sequences with the same length. However, in robotics, this is rarely the case, generalization in HRC is a desirable characteristic that indicates the recorded executions would have different lengths for a specific movement. To address this problem, we first extend the concept of stacked LSTMs to predict anomalies by admitting the input multivariate time series of varying lengths. Subsequently, prediction errors with different lengths are modeled using a probabilistic model for tackling the temporal uncertainty. Consequently, dynamic threshold representation is learned from the trained probabilistic model for abnormal movement identification. A self-designed robot manipulation task consisting of six individual movements is used to evaluate the proposed approach and compared it to baselines. Experimental results indicate that the proposed method with an average anomaly identification accuracy of 94% outperforms the baselines. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Intelligent & Robotic Systems is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)