Intelligent control systems with generalizable behaviour from learned primitives (INTELICOS), 382'049 RON (78'500 EUR), national research grant Young Teams (TE), financed by the Executive Agency for Higher Education, Research, Development and Innovation Funding - UEFISCDI, 2020-2022, project code: PN-III-P1-1.1-TE-2019-1089
The team:
Abstract:
- The project proposal aims at the continuous development of an hierarchical primitives-based learning concept for intelligent control systems (CSs). The idea is to induce feedback CSs with a generalization capability towards tracking tasks, inspired by intelligent living beings who can extrapolate learned optimal behavior to new unseen tasks without learning by repetitions. The framework operates on three levels: L1) low level feedback CS design in model-free data-driven manner to ensure reference tracking, disturbance rejection and indirect CS linearization;L2) learning tracking tasks (in terms of CS reference input + controlled output pairs, called primitives) by repeted executions via model-free data-driven Iterative Learning Control (ILC), over the feedback CS, in terms of a given optimal criterion; L3) extrapolate the learned optimal tracking behavior to new tracking tasks, without needing repetitions. To make the above framework impactful, improvement is needed to: a) ensure strong CS linearization at lower level, in an output reference model tracking problem setting, since the generalizability of the learned tracking bevahvior relies on the superposition principle of the linear CS; b) ensure learning convergence at level L2 via ILC, while reducing the number of dedicated gradient experiments; c) deal with tracking tasks of different time length (shorter/longer) than that of the learned primtives and with operational constraints. The project’s main goals are: to improve the issues a), b), c) and to experimentally validate the hierarchical learning framework on different processes of different nature (tracking tasks are ubiquitous); publish the results in visible journals and conferences; solve the project management issues.
Estimated results:
- The publication of papers in leading journals, participation and presentation of papers in international academic conferences, three scientific reports (two intermediate and a final one).
Obtained results / Rezultate obținute:
- Proiectul și-a indeplinit în totalitate obiectivele propuse. Publicarea rezultatelor a depășit indicatorii de rezultat propuși în contractul de finanțare. Soluția de învățare ierarhizată a fost validată la toate nivelurile, pe foarte multe probleme de urmărire a traiectoriei, pe echipamente de naturi diferite, confirmând omniprezența și importanța aplicației. Validarea experimentală a fost efectuată pentru traiectorii de lungimi diferite, cu restricții asupra ieșirii, cu garantarea convergenței învățării prin reutilizarea datelor experimentale, cu validare pe echipamente neliniare mono- și multivariabile cu importanță industrială foarte relevantă (reamintim aici, sisteme electrice și mecatronice, sisteme de control al mișcării cu video-feedback, sisteme cu suspensie activă, sistem termic, brațe robotice). Considerăm că prin diversitatea aplicațiilor abordate și prin prisma validării cu succes a cadrului de învățare propus, am demonstrat că sistemele de reglare automată pot fi înzestrate cu proprietăți care sunt caracteristice ființelor inteligente: memorarea și învățarea din experiențele anterioare, extrapolarea/generalizarea comportamentului de urmărire învățat în anumite scenarii la scenarii noi nemaivăzute și pentru care nu mai este necesară perfecționarea urmăririi prin repetiții. De asemenea, sistemele de reglare rezultate evidențiază adaptabilitate și robustețe în contextul unor perturbații de sarcină sau de tip zgomot, dar și în fața unor incertitudini de modelare și respectiv în absența liniarizării perfecte la nivelul de reglare de bază. În plus, proiectul a propus o tehnica nouă de reglare după stare virtuală (vezi VSFRT) ca alternativă viabilă și competitivă la regulatoarele după stare virtuală învățate prin întărire cu Value Iteration.
Research reports:
Overall results (2020-2022):
- 3 research reports (2 intermediate, 1 final),
- 6 papers published in Web of Science journals, 5 with impact factors, cumulated impact factors at publication time, according to Journal Citation Reports = 14.749.
- 4 papers published in indexed conferences (IEEE, or Web of Science).
- 1 book chapter published with Springer
Results in 2022
- Radac M.-B., Trajectory Tracking within a Hierarchical Primitive-Based Learning Approach, Entropy, vol. 24, no 7, id:889, 2022, impact factor (IF) at publication = 2.738 (link).
- Lala T., Chirla D.-P., Radac M.-B., Model Reference Tracking Control Solutions for a Visual Servo System Based on a Virtual State from Unknown Dynamics, Energies, vol. 15, no 1., id:267, 2022, impact factor (IF) at publication = 3.004 (link).
- Radac M.-B. and Borlea, A.-B., "Learning Model-Free Reference Tracking Control with Affordable Systems" in Intelligent Techniques for Efficient Use of Valuable Resources - Knowledge and Cultural Resources, L. Ivascu et al. (eds), Springer Book Series, 2022, pp. 1-26. (link).
- Borlea, A.-B. and M.-B. Radac. "A hierarchical learning framework for generalizing tracking control behavior of a laboratory electrical system" in Proc. 2022 IEEE 17th International Conference on Control & Automation (ICCA) June 27-30, 2022 (Hybrid) Naples, Italy, pp 231-236. (link).
- Borlea, A.-B., Lala, T., and Radac, M.-B. 2022b. "A hierarchical learning approach for generalized trajectory tracking validated on a magnetic bearing system." in Proc. International Conference on Electrical, Computer and Energy Technologies (ICECET 2022), 20-22 July 2022, Prague-Czech Republic. (link).
Results in 2021
- 1 intermediate research report.
- Radac M.-B., Lala T., Hierarchical Cognitive Control for Unknown Dynamic Systems Tracking, Mathematics, vol. 9, no 21., id:2752, 2021, impact factor (IF) at publication = 2.258 (link).
- Radac M.-B., Lala, T., A Hierarchical Primitive-Based Learning Tracking Framework for Unknown Observable Systems Based on a New State Representation, in Proc. 2021 European Control Conference (ECC), Rotterdam, Netherlands, 2021, pp. 1466-1472. (link).
- Lala, T., Radac M.-B., Learning to extrapolate an optimal tracking control behavior towards new tracking tasks in a hierarchical primitive-based framework, in Proc. 2021 29th Mediterranean Conference on Control and Automation (MED), Bari, Italy, 2021, pp. 421-427. (link).
- Radac M.-B., Borlea A.-I., Virtual State Feedback Reference Tuning and Value Iteration Reinforcement Learning for Unknown Observable Systems Control, Energies, vol. 14, no 4., id:1006, 2021, impact factor (IF) at publication = 3.004 (link).
- Chereji E., Radac M.-B., Szedlak-Stinean A.-I., Sliding Mode Control Algorithms for Anti-Lock Braking Systems with Performance Comparisons, Algorithms, vol. 14, no. 1, id:2, 2021 (link).
- Radac M.-B., Lala T., Robust Control of Unknown Observable Nonlinear Systems Solved as a Zero-Sum Game, IEEE Access, vol. 8, pp. 214153-214165, 2020, impact factor (IF) at publication = 3.745 ( ieeexplore.ieee.org).
Results in 2020:
- 1 intermediate research report.
Home