{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,25]],"date-time":"2026-03-25T15:53:48Z","timestamp":1774454028424,"version":"3.50.1"},"reference-count":41,"publisher":"Springer Science and Business Media LLC","issue":"1","license":[{"start":{"date-parts":[[2026,1,17]],"date-time":"2026-01-17T00:00:00Z","timestamp":1768608000000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2026,1,17]],"date-time":"2026-01-17T00:00:00Z","timestamp":1768608000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"funder":[{"DOI":"10.13039\/100005825","name":"National Institute of Food and Agriculture","doi-asserted-by":"publisher","award":["2022-67021-37868"],"award-info":[{"award-number":["2022-67021-37868"]}],"id":[{"id":"10.13039\/100005825","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["Auton Robot"],"published-print":{"date-parts":[[2026,3]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>\n                    Robots should be able to learn complex behaviors from human demonstrations. In practice, these human-provided datasets are inevitably\n                    <jats:italic>imbalanced<\/jats:italic>\n                    : i.e., the human demonstrates some subtasks more frequently than others. State-of-the-art methods default to treating each element of the human\u2019s dataset as equally important. So if\u2014for instance\u2014the majority of the human\u2019s data focuses on reaching a goal, and only a few state-action pairs move to avoid an obstacle, the learning algorithm will place greater emphasis on goal reaching. More generally, misalignment between the relative amounts of data and the importance of that data causes fundamental problems for imitation learning approaches. In this paper we analyze and develop learning methods that automatically account for mixed datasets. We formally prove that imbalanced data leads to imbalanced policies when each state-action pair is weighted equally; these policies emulate the most represented behaviors, and not the human\u2019s complex, multi-task demonstrations. We next explore algorithms that rebalance offline datasets (i.e., reweight the importance of different state-action pairs) without human oversight. Reweighting the dataset can enhance the overall policy performance. However, there is no free lunch: each method for autonomously rebalancing brings its own pros and cons. We formulate these advantages and disadvantages, helping other researchers identify when each type of approach is most appropriate. We conclude by introducing a novel meta-gradient rebalancing algorithm that addresses the primary limitations behind existing approaches. Our experiments show that dataset rebalancing leads to better downstream learning, improving the performance of general imitation learning algorithms without requiring additional data collection. See our project website:\n                    <jats:ext-link xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" xlink:href=\"https:\/\/collab.me.vt.edu\/data_curation\/\" ext-link-type=\"uri\">https:\/\/collab.me.vt.edu\/data_curation\/<\/jats:ext-link>\n                    .\n                  <\/jats:p>","DOI":"10.1007\/s10514-025-10237-0","type":"journal-article","created":{"date-parts":[[2026,1,17]],"date-time":"2026-01-17T06:42:16Z","timestamp":1768632136000},"update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Towards balanced behavior cloning from imbalanced datasets"],"prefix":"10.1007","volume":"50","author":[{"given":"Sagar","family":"Parekh","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Heramb","family":"Nemlekar","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Dylan P.","family":"Losey","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"297","published-online":{"date-parts":[[2026,1,17]]},"reference":[{"key":"10237_CR1","unstructured":"Agia, C., Sinha, R., Yang, J., Antonova, R., Pavone, M., Nishimura, H., Itkina, M., & Bohg, J. (2025). Cupid: Curating data your robot loves with influence functions. arXiv preprint arXiv:2506.19121"},{"key":"10237_CR2","doi-asserted-by":"publisher","first-page":"13686","DOI":"10.1109\/ACCESS.2025.3531662","volume":"13","author":"M Altalhan","year":"2025","unstructured":"Altalhan, M., Algarni, A., & Alouane, M. T. H. (2025). Imbalanced data problem in machine learning: A review. IEEE Access, 13, 13686\u201313699.","journal-title":"IEEE Access"},{"key":"10237_CR3","doi-asserted-by":"crossref","unstructured":"Argall, B. D., Browning, B., & Veloso, M. (2009). Automatic weight learning for multiple data sources when learning from demonstration. In IEEE international conference on robotics and automation (pp. 226\u2013231).","DOI":"10.1109\/ROBOT.2009.5152668"},{"key":"10237_CR4","doi-asserted-by":"publisher","first-page":"80375","DOI":"10.52202\/075280-3524","volume":"36","author":"S Belkhale","year":"2023","unstructured":"Belkhale, S., Cui, Y., & Sadigh, D. (2023). Data quality in imitation learning. Advances in Neural Information Processing Systems, 36, 80375\u201380395.","journal-title":"Advances in Neural Information Processing Systems"},{"key":"10237_CR5","doi-asserted-by":"crossref","unstructured":"Chen, A. S., Lessing, A. M., Liu, Y., & Finn, C. (2025). Curating demonstrations using online experience. arXiv preprint arXiv:2503.03707","DOI":"10.15607\/RSS.2025.XXI.071"},{"key":"10237_CR6","doi-asserted-by":"publisher","first-page":"6390","DOI":"10.1109\/TNNLS.2021.3136503","volume":"34","author":"D Dablain","year":"2022","unstructured":"Dablain, D., Krawczyk, B., & Chawla, N. V. (2022). Deepsmote: Fusing deep learning and smote for imbalanced data. IEEE Transactions on Neural Networks and Learning Systems, 34, 6390\u20136404.","journal-title":"IEEE Transactions on Neural Networks and Learning Systems"},{"key":"10237_CR7","unstructured":"Dai, Y., Sanchez, R. R., Jeronimus, R., Sagheb, S., Nunez, C. M., Nemlekar, H., & Losey, D. P. (2025). Civil: Causal and intuitive visual imitation learning. arXiv preprint arXiv:2504.17959"},{"key":"10237_CR8","unstructured":"Dass, S., Khaddaj, A., Engstrom, L., Madry, A., Ilyas, A., & Mart\u00edn-Mart\u00edn, R. (2025). Datamil: Selecting data for robot imitation learning with datamodels. arXiv preprint arXiv:2505.09603"},{"key":"10237_CR9","doi-asserted-by":"crossref","unstructured":"Devi, D., Biswas, S. K., & Purkayastha, B. (2020). A review on solution to class imbalance problem: Undersampling approaches. In 2020 international conference on computational performance evaluation (ComPE) (pp. 626\u2013631).","DOI":"10.1109\/ComPE49325.2020.9200087"},{"key":"10237_CR10","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1016\/j.ins.2018.06.056","volume":"465","author":"G Douzas","year":"2018","unstructured":"Douzas, G., Bacao, F., & Last, F. (2018). Improving imbalanced learning through a heuristic oversampling method based on k-means and smote. Information Sciences, 465, 1\u201320.","journal-title":"Information Sciences"},{"key":"10237_CR11","doi-asserted-by":"crossref","unstructured":"Du, M., Nair, S., Sadigh, D., & Finn, C. (2023). Behavior retrieval: Few-shot imitation learning by querying unlabeled datasets. arXiv preprint arXiv:2304.08742","DOI":"10.15607\/RSS.2023.XIX.011"},{"key":"10237_CR12","doi-asserted-by":"crossref","unstructured":"Fern\u00e1ndez, A., Garc\u00eda, S., Galar, M., Prati, R. C., Krawczyk, B., & Herrera, F. (2018). Learning from imbalanced data sets 10.","DOI":"10.1007\/978-3-319-98074-4"},{"key":"10237_CR13","doi-asserted-by":"publisher","first-page":"863","DOI":"10.1613\/jair.1.11192","volume":"61","author":"A Fern\u00e1ndez","year":"2018","unstructured":"Fern\u00e1ndez, A., Garcia, S., Herrera, F., & Chawla, N. V. (2018). Smote for learning from imbalanced data: Progress and challenges. Journal of Artificial Intelligence Research, 61, 863\u2013905.","journal-title":"Journal of Artificial Intelligence Research"},{"key":"10237_CR14","first-page":"60048","volume":"36","author":"H Fu","year":"2023","unstructured":"Fu, H., Tang, K., Lu, Y., Qi, Y., Deng, G., Sung, F., & Chen, C. (2023). Ess-infogail: Semi-supervised imitation learning from imbalanced demonstrations. Advances in Neural Information Processing Systems, 36, 60048\u201360059.","journal-title":"Advances in Neural Information Processing Systems"},{"key":"10237_CR15","unstructured":"Gandhi, K., Karamcheti, S., Liao, M., Sadigh, & D. (2023). Eliciting compatible demonstrations for multi-human imitation learning. In Conference on robot learning (pp. 1981\u20131991)."},{"key":"10237_CR16","doi-asserted-by":"crossref","unstructured":"Gavenski, N., Monteiro, J., Medronha, A., & Barros, R. C. (2022). How resilient are imitation learning methods to sub-optimal experts? In Brazilian conference on intelligent systems (pp. 449\u2013463).","DOI":"10.1007\/978-3-031-21689-3_32"},{"key":"10237_CR17","unstructured":"Hejna, J., Bhateja, C., Jian, Y., Pertsch, K., & Sadigh, D. (2024). Re-Mix: Optimizing data mixtures for large scale imitation learning. arXiv preprint arXiv:2408.14037"},{"key":"10237_CR18","doi-asserted-by":"crossref","unstructured":"Hejna, J., Mirchandani, S., Balakrishna, A., Xie, A., Wahid, A., Tompson, J., Sanketi, P., Shah, D., Devin, C., & Sadigh, D. (2025). Robot data curation with mutual information estimators. arXiv preprint arXiv:2502.08623","DOI":"10.15607\/RSS.2025.XXI.023"},{"key":"10237_CR19","doi-asserted-by":"crossref","unstructured":"Hou, M., Hindriks, K., Eiben, A., & Baraka, K. (2023). Shaping imbalance into balance: Active robot guidance of human teachers for better learning from demonstrations. In IEEE international conference on robot and human interactive communication (RO-MAN) (pp. 1737\u20131744).","DOI":"10.1109\/RO-MAN57019.2023.10309481"},{"key":"10237_CR20","first-page":"74921","volume":"36","author":"Y Jang","year":"2024","unstructured":"Jang, Y., Kim, G. H., Lee, J., Sohn, S., Kim, B., Lee, H., & Lee, M. (2024). Safedice: Offline safe imitation learning with non-preferred demonstrations. Advances in Neural Information Processing Systems, 36, 74921\u201374951.","journal-title":"Advances in Neural Information Processing Systems"},{"key":"10237_CR21","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1186\/s40537-019-0192-5","volume":"6","author":"JM Johnson","year":"2019","unstructured":"Johnson, J. M., & Khoshgoftaar, T. M. (2019). Survey on deep learning with class imbalance. Journal of Big Data, 6, 1\u201354.","journal-title":"Journal of Big Data"},{"key":"10237_CR22","doi-asserted-by":"publisher","first-page":"3573","DOI":"10.1109\/TNNLS.2017.2732482","volume":"29","author":"SH Khan","year":"2018","unstructured":"Khan, S. H., Hayat, M., Bennamoun, M., Sohel, F. A., & Togneri, R. (2018). Cost-sensitive learning of deep feature representations from imbalanced data. IEEE Transactions on Neural Networks and Learning Systems, 29, 3573\u20133587.","journal-title":"IEEE Transactions on Neural Networks and Learning Systems"},{"key":"10237_CR23","doi-asserted-by":"crossref","unstructured":"Khazatsky, A., Pertsch, K., Nair, S., Balakrishna, A., Dasari, S., Karamcheti, S., Nasiriany, S., Srirama, M.K., Chen, L. Y., & Ellis, K. (2024). Droid: A large-scale in-the-wild robot manipulation dataset. arXiv preprint arXiv:2403.12945","DOI":"10.15607\/RSS.2024.XX.120"},{"key":"10237_CR24","unstructured":"Kim, M. J., Pertsch, K., Karamcheti, S., Xiao, T., Balakrishna, A., Nair, S., Rafailov, R., Foster, E., Lam, G., & Sanketi, P. (2024). Openvla: An open-source vision-language-action model. arXiv preprint arXiv:2406.09246"},{"key":"10237_CR25","doi-asserted-by":"publisher","DOI":"10.1016\/j.patcog.2020.107262","volume":"102","author":"M Koziarski","year":"2020","unstructured":"Koziarski, M. (2020). Radial-based undersampling for imbalanced data classification. Pattern Recognition, 102, Article 107262.","journal-title":"Pattern Recognition"},{"key":"10237_CR26","doi-asserted-by":"publisher","first-page":"17","DOI":"10.1016\/j.ins.2017.05.008","volume":"409","author":"WC Lin","year":"2017","unstructured":"Lin, W. C., Tsai, C. F., Hu, Y. H., & Jhang, J. S. (2017). Clustering-based undersampling in class-imbalanced data. Information Sciences, 409, 17\u201326.","journal-title":"Information Sciences"},{"key":"10237_CR27","doi-asserted-by":"crossref","unstructured":"Mees, O., Hermann, L., Rosete-Beas, E., & Burgard, W. (2022). Calvin: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks. IEEE Robotics and Automation Letters (RA-L) 7, 7327\u20137334.","DOI":"10.1109\/LRA.2022.3180108"},{"key":"10237_CR28","unstructured":"Obuchi, T., & Tanaka, T. (2024). When resampling\/reweighting improves feature learning in imbalanced classification? A toy-model study. arXiv preprint arXiv:2409.05598"},{"key":"10237_CR29","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1561\/2300000053","volume":"7","author":"T Osa","year":"2018","unstructured":"Osa, T., Pajarinen, J., Neumann, G., Bagnell, J. A., Abbeel, P., Peters, J., et al. (2018). An algorithmic perspective on imitation learning. Foundations and Trends in Robotics, 7, 1\u2013179.","journal-title":"Foundations and Trends in Robotics"},{"key":"10237_CR30","doi-asserted-by":"crossref","unstructured":"O\u2019Neill, A., Rehman, A., Maddukuri, A., Gupta, A., Padalkar, A., Lee, A., Pooley, A., Gupta, A., Mandlekar, A., & Jain, A. (2024). Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collaboration 0. In IEEE international conference on robotics and automation (ICRA) (pp. 6892\u20136903).","DOI":"10.1109\/ICRA57147.2024.10611477"},{"key":"10237_CR31","doi-asserted-by":"crossref","unstructured":"Reuss, M., Ya\u011fmurlu, \u00d6.E., Wenzel, F., & Lioutikov, R. (2024). Multimodal diffusion transformer: Learning versatile behavior from multimodal goals. In Robotics: Science and systems.","DOI":"10.15607\/RSS.2024.XX.121"},{"key":"10237_CR32","unstructured":"Sagawa, S., Koh, P. W., Hashimoto, T. B., & Liang, P. (2019). Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization. arXiv preprint arXiv:1911.08731"},{"key":"10237_CR33","unstructured":"Sagheb, S., & Losey, D. P. (2025). Counterfactual behavior cloning: Offline imitation learning from imperfect human demonstrations. arXiv preprint arXiv:2505.10760"},{"key":"10237_CR34","doi-asserted-by":"crossref","unstructured":"Samadi, A., Koufos, K., Debattista, K., & Dianati, M. (2024). Good data is all imitation learning needs. In European conference on computer vision (pp. 144\u2013159).","DOI":"10.1007\/978-3-031-91767-7_10"},{"key":"10237_CR35","unstructured":"Shi, M., Chen, L., Chen, J., Lu, Y., Liu, C., Ren, G., Luo, P., Huang, D., Yao, M., & Li, H. (2025). Is diversity all you need for scalable robotic manipulation? arXiv preprint arXiv:2507.06219"},{"key":"10237_CR36","unstructured":"Team, O. M., Ghosh, D., Walke, H., Pertsch, K., Black, K., Mees, O., Dasari, S., Hejna, J., Kreiman, T., & Xu, C. (2024). Octo: An open-source generalist robot policy. arXiv preprint arXiv:2405.12213"},{"key":"10237_CR37","unstructured":"Wang, Q., Xiong, J., Han, L., Liu, H., & Zhang, T. (2018). Exponentially weighted imitation learning for batched historical data. Advances in Neural Information Processing Systems 31"},{"key":"10237_CR38","unstructured":"Xu, H., Zhan, X., Yin, H., & Qin, H. (2022). Discriminator-weighted offline imitation learning from suboptimal demonstrations. In International conference on machine learning (pp. 24725\u201324742)."},{"key":"10237_CR39","doi-asserted-by":"publisher","first-page":"7173","DOI":"10.1109\/TCYB.2024.3395626","volume":"54","author":"M Zare","year":"2024","unstructured":"Zare, M., Kebria, P. M., Khosravi, A., & Nahavandi, S. (2024). A survey of imitation learning: Algorithms, recent developments, and challenges. IEEE Transactions on Cybernetics, 54, 7173\u20137186.","journal-title":"IEEE Transactions on Cybernetics"},{"key":"10237_CR40","unstructured":"Zha, L., Badithela, A., Zhang, M., Lidard, J., Bao, J., Zhou, E., Snyder, D., Ren, A. Z., Shah, D., & Majumdar, A. (2025). Guiding data collection via factored scaling curves. arXiv preprint arXiv:2505.07728"},{"key":"10237_CR41","unstructured":"Zhang, W., Xu, H., Niu, H., Cheng, P., Li, M., Zhang, H., Zhou, G., & Zhan, X. (2023). Discriminator-guided model-based offline imitation learning. In Conference on robot learning (pp. 1266\u20131276)."}],"container-title":["Autonomous Robots"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/link.springer.com\/content\/pdf\/10.1007\/s10514-025-10237-0.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/link.springer.com\/article\/10.1007\/s10514-025-10237-0","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/link.springer.com\/content\/pdf\/10.1007\/s10514-025-10237-0.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,25]],"date-time":"2026-03-25T15:02:40Z","timestamp":1774450960000},"score":1,"resource":{"primary":{"URL":"https:\/\/link.springer.com\/10.1007\/s10514-025-10237-0"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,1,17]]},"references-count":41,"journal-issue":{"issue":"1","published-print":{"date-parts":[[2026,3]]}},"alternative-id":["10237"],"URL":"https:\/\/doi.org\/10.1007\/s10514-025-10237-0","relation":{},"ISSN":["0929-5593","1573-7527"],"issn-type":[{"value":"0929-5593","type":"print"},{"value":"1573-7527","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,1,17]]},"assertion":[{"value":"7 August 2025","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"6 November 2025","order":2,"name":"revised","label":"Revised","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"31 December 2025","order":3,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"17 January 2026","order":4,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"order":1,"name":"Ethics","group":{"name":"EthicsHeading","label":"Declarations"}},{"value":"The authors declare that they have no Conflict of interest.","order":2,"name":"Ethics","group":{"name":"EthicsHeading","label":"Conflict of interest"}}],"article-number":"9"}}