Robust Task Generalization for Dual-Arm Learning from Demonstration
DOI:
https://doi.org/10.17979/ja-cea.2026.47.13706Palabras clave:
Aprendizaje por demostración, Manipulación bibrazo, Aprendizaje adaptativoResumen
La manipulación bibrazo o la coordinación física humano-robot, requieren que los robots se adapten rápidamente a entornos y restricciones cambiantes. Los enfoques tradicionales de Aprendizaje por Demostración tienen dificultades para generalizar frente a escenarios fuera de distribución, requiriendo costosos reentrenamientos. Proponemos un algoritmo de aprendizaje de Primitivas de Movimiento basado en Procesos Gaussianos, combinado con una adaptación zero-shot en tiempo real mediante Pathwise Conditioning. El método encapsula la incertidumbre predictiva del movimiento demostrado mediante GPs heterocedásticos y utiliza una actualización mediante la regla de Matheron para ajustar la trayectoria a nuevos via-points de manera instantánea, sin necesidad de reentrenar el modelo subyacente. Esta formulación se extiende a la coordinación bimanual calculando dinámicamente restricciones relativas en 6D para mantener una cadena cinemática cerrada. Los resultados experimentales, en comparativas 2D contra modelos parametrizados por tareas y en tareas con el robot ADAM, demuestran una adaptación robusta con un error cercano a cero y en tiempo real, pudiendo aplicarse para entornos altamente cambiantes.
Referencias
Arduengo, M., Colomé, A., Borras, J., Torras, C., 2021. Task-adaptive robot learning from demonstration with gaussian process models under replication. IEEE RAL 6 (2), 966–973.
Barber, R., Ortiz, F. J., Garrido, S., Mora, A., Prados, A., Méndez, I., Mozos, Ó . M., 2022. A multirobot system in an assisted home environment to support the elderly in their daily lives. Sensors 22 (20), 7983.
Calinon, S., 2016. A tutorial on task-parameterized movement learning and retrieval. Intelligent service robotics 9 (1), 1–29.
Calinon, S., Caldwell, D. G., 2014. A task-parameterized probabilistic model with minimal intervention control. In: ICRA. IEEE, pp. 39–44.
Carrasco, A. P., Velasco, A. M., García, A. M., Bullón, S. G., Castaño, R. I. B., 2024. Learning from demonstration through synthetic data for parameterized tasks. Jornadas de Autom´atica (45).
Feng, C., Liu, Z., Li, W., Lu, X., Jing, Y., Ma, Y., 2025. Improved gaussian mixture model and gaussian mixture regression for learning from demonstration based on gaussian noise scattering. Adv. Eng. Infor. 65, 103192.
Huang, K., Ji, X., Su, J., Qu, X., 2025. Task-parameterized dynamic movement primitives with reinforcement learning for improved motion planning. IEEE Robotics and Automation Letters.
Lee, Q. Y., Kulkarni, S. R., Yang, L., Noronha, B., Wee, Y., Campolo, D., 2025. Generalizing robot trajectories from single-context human demonstrations: A probabilistic approach. arXiv preprint arXiv:2503.05619.
Li, T., Figueroa, N., 2023. Task generalization with stability guarantees via elastic dynamical system motion policies. In: CORL.
Mehta, S. A., Ciftci, Y. U., Ramachandran, B., Bansal, S., Losey, D. P., 2025. Stable-bc: Controlling covariate shift with stable behavior cloning. IEEE Robotics and Automation Letters.
Mendez, A., Prados, A., Menendez, E., Barber, R., 2024. Everyday objects rearrangement in a human-like manner via robotic imagination and learning from demonstration. IEEE Access 12, 98–119.
Mora, A., Prados, A., Mendez, A., Espinoza, G., Gonzalez, P., Lopez, B., Muñoz, V., Moreno, L., Garrido, S., Barber, R., 2024. Adam: a robotic companion for enhanced quality of life in aging populations. Frontiers in Neurorobotics 18, 1337608.
Paraschos, A., Daniel, C., Peters, J. R., Neumann, G., 2013. Probabilistic movement primitives. Adv. in neural info. processing 26.
Prados, A., Espinoza, G., Mendez, A., Mora, A., Garrido, S., Barber, R., 2025a. Adamsim: Pybullet-based simulation environment for research on domestic mobile manipulator robots. Jornadas de Auto. (46).
Prados, A., Espinoza, G., Moreno, L., Barber, R., 2025b. Coordination of learned decoupled dual-arm tasks through gaussian belief propagation. In: 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, pp. 15917–15924.
Prados, A., Garrido, S., Barber, R., 2024a. Learning and generalization of task-parameterized skills through few human demonstrations. Eng. Applications of Artificial Intelligence 133, 108310.
Prados, A., Hertel, B., Barber, R., Azadeh, R., 2026a. Elastic fast marching learning from demonstration. Advanced Intelligent Systems 8 (2), e202500607.
Prados, A., Mendez, A., Espinoza, G., Fernandez, N., Barber, R., 2024b. fdivergence optimization for task-parameterized learning from demonstrations algorithm. In: IEEE Int. Conference on Autonomous Robot Systems and Competitions (ICARSC). IEEE, pp. 9–14.
Prados, A., Mora, A., López, B., Muñoz, J., Garrido, S., Barber, R., 2023. Kinesthetic learning based on fast marching square method for manipulation. Applied Sciences 13 (4), 2028.
Prados, A., Moreno, L., Barber, R., 2026b. Learning of movement primitives by gaussian processes from demonstrations. Engineering Applications of Artificial Intelligence 179, 115258.
Qian, K., Xu, X., Liu, H., Bai, J., Luo, S., 2022. Environment-adaptive learning from demonstration for proactive assistance in human–robot collaborative tasks. Robotics and Autonomous Systems 151, 104046.
Ruan, S., Meng, X., Chirikjian, G. S., 2024. Primp: Probabilistically informed motion primitives for efficient affordance learning from demonstration. IEEE Trans. on Rob. 40, 68–87.
Silvério, J., Rozo, L., Calinon, S., Caldwell, D. G., 2015. Learning bimanual end-effector poses from demonstrations using task-parameterized dynamical systems. In: 2015 IROS. IEEE, pp. 464–470.
Sosa-Ceron, A. D., Gonzalez-Hernandez, H. G., Reyes-Avendaño, J. A., 2022. Learning from demonstrations in human–robot collaborative scenarios: A survey. Robotics 11 (6), 126.
Spencer, J., Choudhury, A., Ziebart, B., Bagnell, J. A., 2021. Feedback in imitation learning: The three regimes of covariate shift. arXiv:2102.02872.
Wilson, J. T., Borovitskiy, V., Terenin, A., Mostowsky, P., Deisenroth, M. P., 2021. Pathwise conditioning of gaussian processes. Journal of Machine Learning Research 22 (105), 1–47.
Zhang, R., Xia, J., Ma, J., Huang, D., Zhang, X., Li, Y., 2024. Human–robot interactive skill learning and correction for polishing based on Dynamic time warping iterative learning control. IEEE Transactions on Control Systems Technology.
Descargas
Publicado
Número
Sección
Licencia
Derechos de autor 2026 Adrian Prados, Lucia Lishan, Alberto Mendez, Santiago Garrido, Ramon Barber
Esta obra está bajo una licencia internacional Creative Commons Atribución-NoComercial-CompartirIgual 4.0.
