
Basic data
- Manufactura
IVACE
Development of a complex trajectory generation system for robots based on learning by demonstration.
IVACE
Currently, one of the biggest problems in industry is the programming of robots to perform tasks in highly changing environments. Traditional programming methods, among other problems, require extensive technical knowledge on the part of the responsible personnel and are time-consuming. From this arises the need to facilitate the process of programming industrial operations in robots, a field in which AIDIMME has been working for several years.
One of the techniques being experimented with is called Learning by Demonstration (LbD). This method consists mainly of "observing" the performance of a task, extracting relevant information and generating actions that can be interpreted and reproduced by the robot controller. To be easily adopted by industry, this conversion process must be fast (without human intervention as far as possible), reliable and accurate, so that the observed tasks are reproduced faithfully and without errors.
With these premises, the project proposes to advance in the development of techniques to speed up robot programming by applying Demonstration Learning techniques based on the results obtained in a previous project (COLEARNING 4.0 - Development of learning techniques for COBOTS based on human interaction and reinforcement learning. IMDEEA/2020/22), in which the robot was able to reproduce movements and pressures previously performed by a person, although in an approximate way and with a manual treatment of the captured data.
To further develop this automatic programming technique, the main objective of the project is to develop a learning-by-demonstration system for collaborative robots, based on an artificial vision system and sensors external to the robot, which can accurately reproduce complex tasks immediately after the human demonstration, using artificial intelligence algorithms for the modeling of trajectories and execution of tasks. The validation of the development achieved would be performed on the reproduction of a process with a high manual workload involving the control of several robot parameters, in addition to the trajectories (speed, force, torque, etc.).
Currently, one of the biggest problems in industry is the programming of robots to perform tasks in highly changing environments. Traditional programming methods, among other problems, require extensive technical knowledge on the part of the responsible personnel and are time-consuming. From this arises the need to facilitate the process of programming industrial operations in robots, a field in which AIDIMME has been working for several years.
One of the techniques being experimented with is Learning by Demonstration (LbD). This method consists mainly of “observing” the performance of a task, extracting relevant information and generating actions that can be interpreted and reproduced by the robot controller. To be easily adopted by industry, this conversion process must be fast (without human intervention as far as possible), reliable and accurate, so that the observed tasks are reproduced faithfully and without errors.
With these premises, the project proposes to advance in the development of techniques to speed up robot programming by applying Demonstration Learning techniques based on the results obtained in a previous project (COLEARNING 4.0 – Development of learning techniques for COBOTS based on human interaction and reinforcement learning. IMDEEA/2020/22), in which the robot was able to reproduce movements and pressures previously performed by a person, although in an approximate way and with a manual treatment of the captured data.
To further develop this automatic programming technique, the main objective of the project is to develop a learning-by-demonstration system for collaborative robots, based on an artificial vision system and sensors external to the robot, which allows to accurately reproduce complex tasks immediately after the human demonstration, using artificial intelligence algorithms for the modeling of trajectories and execution of tasks. The validation of the development achieved would be performed on the reproduction of a process with a high manual workload involving the control of several robot parameters, in addition to the trajectories (speed, force, torque, etc.).
José Luis Sánchez Asins
Industrial Development Manager
AIDIMME