"COLEARNING 20-21" COLEARNING-DEVELOPMENT OF TASK LEARNING TECHNIQUES FOR COBOTS BASED ON HUMAN INTERACTION AND REINFORCEMENT LEARNING ALGORITHMS (COLEARNING) DESCRIPTION Generally speaking, robot programming has been greatly simplified in recent years. Manufacturers have developed graphic tools and gestural methods that facilitate the generation of programs in any type of robot. Obviously, each program is valid in a specific environment and for a specific task, so reprogramming is required if either of the two circumstances changes. Therefore, one of the key challenges of deploying COBOTS (collaborative robots) in environments where human presence is important, such as manufacturing industries, is adapting to tasks that change frequently and unpredictably. In these circumstances, the robot should be able to quickly update the way it performs its task. In other words, they should be able to learn incrementally and adapt to the circumstances of both the environment and the task. Beyond the programming systems that each robot manufacturer uses, which tend to simplify and facilitate programming, techniques can be implemented that accelerate robot learning to perform new tasks. And something that would be more important in certain types of work: they would allow facing tasks that are not exactly the same as the one learned, being able to assume variations that are part of the changing nature of the processed products, adapting the work cycle to the characteristics of each product individual. OBJECTIVES The objective of the project is to develop a learning system for COBOTS that allows the reconfiguration of the program autonomously, depending on the variations of the process or product that cannot be fully considered in a standard programming routine. As a use case, it would be applied to a process with a lot of manual workload: the sanding / polishing of pieces of board and / or metal sheet, whose geometry differs slightly from one piece to another due to the forming process or due to the flexibility of the piece itself. material. In this project, two different learning techniques will be used and compared to achieve the objective of self-reconfiguration of the work cycle: on the one hand, human supervision, through which COBOT can acquire new capabilities by demonstrating tasks; on the other hand, reinforcement learning through neural networks and the use of reward functions, which would allow to improve the performance of the task from the results that are obtained. Project Number: 22000052 Grant Agreement: IMDEEA / 2020/22 Duration: From 01/04/2020 to 31/07/2021 Coordinated in AIDIMME by: SÁNCHEZ ASINS, JOSÉ LUIS R&D Line: INDUSTRY 4.0 RESULTS OBTAINED Year 2021: The learning system has been developed based on the imitation of the activity of an expert person, applied to a sanding process. The system is based on the monitoring of manual trajectories and the continuous measurement of the pressures exerted. The reinforcement learning system has been developed, based on a reward function based on two factors: surface roughness and subjective judgment. Systems for capturing pressures (sensored glove) and capturing trajectories (camera and tracking software) of the movements of the expert have been developed. Likewise, trajectory preprocessing tools have been developed, and an application has been generated to download the data to the robot controller. Development of a pilot project in collaborating companies to evaluate the results achieved. Deliverables:    PUBLISHED NEWS general broadcast https://actualidad.aidimme.es/2021/07/26/finalizan-proye (...) https://actualidad.aidimme.es/2022/04/19/mejora-eficienc (...) https://www.interempresas.net/Robotica/Articulos/385646- (...) https://actualidad.aidimme.es/2021/06/02/reuniones-para- (...) https://actualidad.aidimme.es/2021/09/08/difusion-y-tran (...) https://actualidad.aidimme.es/2020/07/16/aidimme-nuevos- (...) https://actualidad.aidimme.es/2020/12/30/especial-difusi (...) Technical dissemination, transfer https://actualidad.aidimme.es/2023/11/29/inteligencia-ar (...) https://actualidad.aidimme.es/2023/09/04/gemelos-digital (...) https://actualidad.aidimme.es/2020/11/17/robots-gemelos- (...) https://actualidad.aidimme.es/2020/09/16/aidimme-sistema (...) https://actualidad.aidimme.es/2021/06/07/aidimme-ha-vali (...) https://actualidad.aidimme.es/2021/03/23/aidimme-desarro (...) GRANT 228.125 € "" TARGET AUDIENCE AND IMPACT MEASUREMENT 5072 Access to the project website 15991 Access to news published on own websites. Total Accesses: 21063 Objective Sectors by CNAE CNAE: 3109 - Manufacture of other furniture - No. of target companies: 5592 (1333 in the Valencian Community) CNAE: 2561 - Treatment and coating of metals - Number of target companies: 107 (63 from the Valencian Community) CNAE: 3101 - Manufacture of office furniture and e - Number of target companies: 575 (115 in the Valencian Community) CNAE: 3103 - Manufacture of mattresses - No. of target companies: 149 (41 in the Valencian Community) CNAE: 3102 - Manufacture of kitchen furniture - Number of target companies: 669 (156 from the Valencian Community) CNAE: 3299 - Other manufacturing industries nec - No. of target companies: 107 (49 from the Valencian Community) Objective Sectors by Activity FURNITURE - Number of target companies: 231 (92 of the Valencian Community) TRACTOR COMPANIES Thanks to its support and signature of the "declaration of participation" the project has been funded. MOTHERSON DRSC MODULES SPAIN, SAU LAGRAMA, SL MUEBLES ROMERO, SA MEDITERRANEAN WOOD FACTORY SLU PUNT MOBLES XXI, SL HURTADO RIVAS, SL COMPANIES DIFFUSION R&D PROJECTS They want to know first-hand the evolution of the project, and its progress to the final result. COMPANIES TRANSFER KNOWLEDGE They will implement technologies, develop strategies or look for new models based on the results.