{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,5,12]],"date-time":"2026-05-12T19:41:10Z","timestamp":1778614870624,"version":"3.51.4"},"reference-count":50,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2023,6,16]],"date-time":"2023-06-16T00:00:00Z","timestamp":1686873600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Neurorobot."],"abstract":"<jats:p>Collaborative robots, or cobots, are designed to work alongside humans and to alleviate their physical burdens, such as lifting heavy objects or performing tedious tasks. Ensuring the safety of human\u2013robot interaction (HRI) is paramount for effective collaboration. To achieve this, it is essential to have a reliable dynamic model of the cobot that enables the implementation of torque control strategies. These strategies aim to achieve accurate motion while minimizing the amount of torque exerted by the robot. However, modeling the complex non-linear dynamics of cobots with elastic actuators poses a challenge for traditional analytical modeling techniques. Instead, cobot dynamic modeling needs to be learned through data-driven approaches, rather than analytical equation-driven modeling. In this study, we propose and evaluate three machine learning (ML) approaches based on bidirectional recurrent neural networks (BRNNs) for learning the inverse dynamic model of a cobot equipped with elastic actuators. We also provide our ML approaches with a representative training dataset of the cobot's joint positions, velocities, and corresponding torque values. The first ML approach uses a non-parametric configuration, while the other two implement semi-parametric configurations. All three ML approaches outperform the rigid-bodied dynamic model provided by the cobot's manufacturer in terms of torque precision while maintaining their generalization capabilities and real-time operation due to the optimized sample dataset size and network dimensions. Despite the similarity in torque estimation of these three configurations, the non-parametric configuration was specifically designed for worst-case scenarios where the robot dynamics are completely unknown. Finally, we validate the applicability of our ML approaches by integrating the worst-case non-parametric configuration as a controller within a feedforward loop. We verify the accuracy of the learned inverse dynamic model by comparing it to the actual cobot performance. Our non-parametric architecture outperforms the robot's default factory position controller in terms of accuracy.<\/jats:p>","DOI":"10.3389\/fnbot.2023.1166911","type":"journal-article","created":{"date-parts":[[2023,6,16]],"date-time":"2023-06-16T05:16:55Z","timestamp":1686892615000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":11,"title":["Bidirectional recurrent learning of inverse dynamic models for robots with elastic joints: a real-time real-world implementation"],"prefix":"10.3389","volume":"17","author":[{"given":"Brayan","family":"Valencia-Vidal","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Eduardo","family":"Ros","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Ignacio","family":"Abad\u00eda","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Niceto R.","family":"Luque","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"1965","published-online":{"date-parts":[[2023,6,16]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"381","DOI":"10.1017\/S0263574700019779","article-title":"Causal inverse dynamics of a flexible hub-arm system through Liapunov's second method","volume":"14","author":"Ata","year":"1996","journal-title":"Robotica"},{"key":"B2","doi-asserted-by":"publisher","first-page":"562","DOI":"10.1162\/neco_a_01260","article-title":"Model-free robust optimal feedback mechanisms of biological motor control","volume":"32","author":"Bian","year":"2020","journal-title":"Neural Comput"},{"key":"B3","doi-asserted-by":"publisher","first-page":"1335","DOI":"10.1007\/978-3-540-30301-5_58","article-title":"\u201cSafety for physical human-robot interaction,\u201d","author":"Bicchi","year":"2008","journal-title":"Springer Handbook of Robotics"},{"key":"B4","doi-asserted-by":"publisher","first-page":"105049","DOI":"10.1016\/j.sysconle.2021.105049","article-title":"On the stability properties of gated recurrent units neural networks","volume":"157","author":"Bonassi","year":"2021","journal-title":"Syst. Control Lett"},{"key":"B5","doi-asserted-by":"publisher","first-page":"11","DOI":"10.1109\/41.121906","article-title":"Analysis of algorithms for velocity estimation from discrete position versus time data","volume":"39","author":"Brown","year":"1992","journal-title":"IEEE Trans. Indus. Electron"},{"key":"B6","doi-asserted-by":"publisher","first-page":"1214","DOI":"10.1109\/TCST.2010.2077294","article-title":"MIMO closed loop identification of an industrial robot","volume":"19","author":"Calanca","year":"2011","journal-title":"IEEE Trans. Control Syst. Technol"},{"key":"B7","doi-asserted-by":"publisher","first-page":"1061","DOI":"10.1109\/LRA.2022.3222951","article-title":"Hybrid learning of time-series inverse dynamics models for locally isotropic robot motion","volume":"8","author":"\u00c7allar","year":"2023","journal-title":"IEEE Robot. Autom. Lett"},{"key":"B8","doi-asserted-by":"publisher","first-page":"544","DOI":"10.1109\/ICRA.2016.7487177","article-title":"\u201cIncremental semiparametric inverse dynamics learning,\u201d","author":"Camoriano","year":"2016","journal-title":"2016 IEEE International Conference on Robotics and Automation (ICRA)"},{"key":"B9","doi-asserted-by":"publisher","first-page":"128","DOI":"10.1109\/IROS40897.2019.8968608","article-title":"\u201cNeural-learning trajectory tracking control of flexible-joint robot manipulators with unknown dynamics,\u201d","author":"Chen","year":"2019","journal-title":"2019 IEEE\/RSJ International Conference on Intelligent Robots and Systems (IROS)"},{"key":"B10","article-title":"Empirical evaluation of gated recurrent neural networks on sequence modeling","author":"Chung","year":"2014","journal-title":"arXiv Preprint. arXiv:1412.3555"},{"key":"B11","unstructured":"CoumansE.\n            BaiY.\n          Pybullet, A Python Module for Physics Simulation for Games, Robotics and Machine Learning"},{"key":"B12","doi-asserted-by":"publisher","DOI":"10.1109\/TePRA.2013.6556344","article-title":"\u201cDeveloping baxter,\u201d","author":"Fitzgerald","year":"2013","journal-title":"IEEE Conference on Technologies for Practical Robot Applications, TePRA"},{"key":"B13","author":"Francis","year":"1995","journal-title":"Optimal Sampled-Data Control Systems"},{"key":"B14","doi-asserted-by":"publisher","first-page":"253","DOI":"10.1007\/s12369-010-0052-0","article-title":"Design principles for safety in human-robot interaction","volume":"2","author":"Giuliani","year":"2010","journal-title":"Int. J. Soc. Robot"},{"key":"B15","doi-asserted-by":"publisher","first-page":"799","DOI":"10.1007\/11550907_126","article-title":"\u201cBidirectional LSTM networks for improved phoneme classification and recognition,\u201d","author":"Graves","year":"2005","journal-title":"Artificial Neural Networks: Formal Models and Their Applications-ICANN 2005"},{"key":"B16","doi-asserted-by":"publisher","first-page":"812","DOI":"10.1109\/TIE.2020.2991997","article-title":"Gru-type larc strategy for precision motion control with accurate tracking error prediction","volume":"68","author":"Hu","year":"2021","journal-title":"IEEE Trans. Indus. Electron"},{"key":"B17","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1109\/TMECH.2021.3076519","article-title":"Robust estimation for an extended dynamic parameter set of serial manipulators and unmodeled dynamics compensation","volume":"4435","author":"Huang","year":"2021","journal-title":"IEEE\/ASME Trans. Mechatron"},{"key":"B18","year":"2021","journal-title":"IFR Presents World Robotics 2021 Reports"},{"key":"B19","doi-asserted-by":"publisher","first-page":"1065256","DOI":"10.3389\/fnbot.2022.1065256","article-title":"A robust zeroing neural network and its applications to dynamic complex matrix equation solving and robotic manipulator trajectory tracking","volume":"16","author":"Jin","year":"2022","journal-title":"Front. Neurorobot"},{"key":"B20","doi-asserted-by":"publisher","first-page":"307","DOI":"10.1207\/s15516709cog1603_1","article-title":"Forward models: supervised learning with a distal teacher","volume":"16","author":"Jordan","year":"1992","journal-title":"Cogn. Sci"},{"key":"B21","doi-asserted-by":"publisher","first-page":"4723","DOI":"10.1109\/IROS.2017.8206345","article-title":"\u201cA new data source for inverse dynamics learning,\u201d","author":"Kappler","year":"2017","journal-title":"2017 IEEE\/RSJ International Conference on Intelligent Robots and Systems (IROS)"},{"key":"B22","article-title":"\u201cAdam: A Method for Stochastic Optimization,\u201d","author":"Kingma","year":"2014","journal-title":"Proceedings of the 3rd International Conference on Learning Representations, San Diego"},{"key":"B23","doi-asserted-by":"publisher","first-page":"193","DOI":"10.1115\/1.2899210","article-title":"A time-domain inverse dynamic tracking control of a single-link flexible manipulator","volume":"116","author":"Kwon","year":"1994","journal-title":"J. Dyn. Syst. Measure. Control"},{"key":"B24","doi-asserted-by":"publisher","first-page":"26","DOI":"10.3390\/act6030026","article-title":"Generalization of Series Elastic Actuator configurations and dynamic behavior comparison","volume":"6","author":"Lee","year":"2017","journal-title":"Actuators"},{"key":"B25","doi-asserted-by":"publisher","first-page":"2854","DOI":"10.1109\/TNNLS.2021.3109953","article-title":"Neural network model-based control for manipulator: An autoencoder perspective","volume":"34","author":"Li","year":"2021","journal-title":"IEEE Trans. Neur. Netw. Learn. Sys."},{"key":"B26","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1007\/978-3-642-01094-1_1","author":"Limon","year":"2009","journal-title":"Input-to-State Stability: A Unifying Framework for Robust Model Predictive Control"},{"key":"B27","doi-asserted-by":"publisher","first-page":"2339","DOI":"10.1109\/TCYB.2020.2978003","article-title":"Neural-network-based sliding-mode control of an uncertain robot using dynamic model approximated switching gain","volume":"51","author":"Liu","year":"2021","journal-title":"IEEE Trans. Cybern"},{"key":"B28","doi-asserted-by":"publisher","first-page":"173989","DOI":"10.1109\/ACCESS.2019.2957019","article-title":"Modeling and simulation of robot inverse dynamics using LSTM-based deep learning algorithm for smart cities and factories","volume":"7","author":"Liu","year":"2019","journal-title":"IEEE Access"},{"key":"B29","doi-asserted-by":"publisher","first-page":"9053715","DOI":"10.1155\/2020\/9053715","article-title":"Semiparametric deep learning manipulator inverse dynamics modeling method for smart city and industrial applications","volume":"2020","author":"Liu","year":"2020","journal-title":"Complexity"},{"key":"B30","doi-asserted-by":"publisher","first-page":"1175","DOI":"10.1016\/j.ifacol.2020.12.1329","article-title":"Deep learning and system identification","volume":"53","author":"Ljung","year":"2020","journal-title":"IFAC-PapersOnLine"},{"key":"B31","doi-asserted-by":"publisher","first-page":"69","DOI":"10.1115\/1.3149599","article-title":"On-line computational scheme for mechanical manipulators","volume":"102","author":"Luh","year":"1980","journal-title":"J. Dyn. Syst. Measure. Control"},{"key":"B32","doi-asserted-by":"publisher","first-page":"104462","DOI":"10.1016\/j.conengprac.2020.104462","article-title":"Comprehensive modeling and identification of nonlinear joint dynamics for collaborative industrial robot manipulators","volume":"101","author":"Madsen","year":"2020","journal-title":"Control Eng. Pract"},{"key":"B33","doi-asserted-by":"publisher","first-page":"3370","DOI":"10.1109\/TNNLS.2019.2891257","article-title":"Multistep prediction of dynamic systems with recurrent neural networks","volume":"30","author":"Mohajerin","year":"2019","journal-title":"IEEE Trans. Neural Netw. Learn. Syst"},{"key":"B34","doi-asserted-by":"publisher","author":"Mozian","year":"2020","DOI":"10.1109\/IROS45743.2020.9341019"},{"key":"B35","doi-asserted-by":"publisher","first-page":"17","DOI":"10.1109\/TENCON.2019.8929622","article-title":"\u201cModel learning for robotic manipulators Usingrecurrent neural networks,\u201d","author":"Mukhopadhyay","year":"2019","journal-title":"Proceedings of the TENCON 2019: Technology, Knowledge, and Society"},{"key":"B36","doi-asserted-by":"publisher","first-page":"380","DOI":"10.1109\/IROS.2008.4650850","article-title":"\u201cLocal Gaussian process regression for real-time model-based robot control,\u201d","author":"Nguyen-Tuong","year":"2008","journal-title":"2008 IEEE\/RSJ International Conference on Intelligent Robots and Systems"},{"key":"B37","doi-asserted-by":"publisher","first-page":"1149","DOI":"10.1109\/TCST.2017.2695600","article-title":"Adaptive command-filtered backstepping control of robot arms with compliant actuators","volume":"26","author":"Pan","year":"2018","journal-title":"IEEE Trans. Control Syst. Technol"},{"key":"B38","doi-asserted-by":"publisher","DOI":"10.1109\/IROS.2015.7353857","author":"Polydoros","year":"2015","journal-title":"Real-Time Deep Learning of Robotic Manipulator Inverse Dynamics"},{"key":"B39","doi-asserted-by":"publisher","first-page":"399","DOI":"10.1109\/IROS.1995.525827","article-title":"\u201cSeries elastic actuators,\u201d","author":"Pratt","year":"1995","journal-title":"Proceedings 1995 IEEE\/RSJ International Conference on Intelligent Robots and Systems. Human Robot Interaction and Cooperative Robots, Vol. 1"},{"key":"B40","doi-asserted-by":"publisher","author":"Rueckert","year":"2017","DOI":"10.1109\/HUMANOIDS.2017.8246965"},{"key":"B41","doi-asserted-by":"publisher","first-page":"448","DOI":"10.1109\/SSCI44817.2019.9003133","author":"Sch\u00fcssler","year":"2019"},{"key":"B42","doi-asserted-by":"publisher","first-page":"2673","DOI":"10.1109\/78.650093","article-title":"Bidirectional recurrent neural networks","volume":"45","author":"Schuster","year":"1997","journal-title":"IEEE Trans. Signal Process"},{"key":"B43","doi-asserted-by":"publisher","first-page":"2039","DOI":"10.1109\/LRA.2020.2970987","article-title":"Online simultaneous semi-parametric dynamics model learning","volume":"5","author":"Smith","year":"2020","journal-title":"IEEE Robot. Autom. Lett"},{"key":"B44","doi-asserted-by":"publisher","first-page":"161","DOI":"10.1016\/j.isatra.2017.09.016","article-title":"Model predictive control for systems with fast dynamics using inverse neural models","volume":"72","author":"Stogiannos","year":"2018","journal-title":"ISA Trans"},{"key":"B45","doi-asserted-by":"publisher","first-page":"58","DOI":"10.1109\/MCS.2007.904659","article-title":"Dynamic model identification for industrial robots","volume":"27","author":"Swevers","year":"2007","journal-title":"IEEE Control Syst. Mag"},{"key":"B46","doi-asserted-by":"publisher","first-page":"329","DOI":"10.1109\/TCYB.2021.3091680","article-title":"A novel exploration-exploitation-based adaptive law for intelligent model-free control approaches","volume":"53","author":"Tutsoy","year":"2023","journal-title":"IEEE Trans. Cybern"},{"key":"B47","doi-asserted-by":"publisher","first-page":"111612","DOI":"10.1016\/j.chaos.2021.111612","article-title":"Dynamical time series embeddings in recurrent neural networks","volume":"154","author":"Uribarri","year":"2022","journal-title":"Chaos Solitons Fract"},{"key":"B48","doi-asserted-by":"publisher","first-page":"138102","DOI":"10.1109\/ACCESS.2020.3012196","article-title":"Deep learning aided dynamic parameter identification of 6-DOF robot manipulators","volume":"8","author":"Wang","year":"2020","journal-title":"IEEE Access"},{"key":"B49","doi-asserted-by":"publisher","first-page":"2626","DOI":"10.1109\/TCYB.2014.2311824","article-title":"Composite neural dynamic surface control of a class of uncertain nonlinear systems in strict-feedback form","volume":"44","author":"Xu","year":"2014","journal-title":"IEEE Trans. Cybern"},{"key":"B50","doi-asserted-by":"publisher","first-page":"98","DOI":"10.1109\/IWECAI50956.2020.00027","article-title":"\u201cLSTM and GRU neural network performance comparison study: taking yelp review dataset as an example,\u201d","author":"Yang","year":"2020","journal-title":"Proceedings"}],"container-title":["Frontiers in Neurorobotics"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fnbot.2023.1166911\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2023,6,16]],"date-time":"2023-06-16T05:17:00Z","timestamp":1686892620000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fnbot.2023.1166911\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2023,6,16]]},"references-count":50,"alternative-id":["10.3389\/fnbot.2023.1166911"],"URL":"https:\/\/doi.org\/10.3389\/fnbot.2023.1166911","relation":{},"ISSN":["1662-5218"],"issn-type":[{"value":"1662-5218","type":"electronic"}],"subject":[],"published":{"date-parts":[[2023,6,16]]},"article-number":"1166911"}}