Learning Intelligible Task Models for Cobots Programming – Prog4Yu

From an industrial point of view, the objective of the prog4Yu project is to develop an industrial cobot demonstrator «programmable by demonstration« to deal with the problem of packaging/palletizing parts at the end of the production line with the YuMi cobot.

From a scientific point of view, the objective is to (1) develop algorithms for learning cobot trajectories (gestures) and intelligible task models (semantics of gestures) according to a demonstration approach and (2) identify the factors that in this approach promote or hinder the acceptance of cobotics in an industrial environment. The operator demonstrations are done by kinesthetic manipulations of the cobot arms and via a multimodal Human-Robot IDE (Integrated Development Environment) interface.

The issues addressed by the project are:

1. Trajectory learning by demonstration: learning trajectories from few demonstrations (kinesthetic manipulation of a robotic arm by the operator), in an incremental and efficient way, i.e. by allowing a fluid interaction with the operator is an important scientific issue that we will have to solve to implement the project's industrial demonstrator.

2. Learning task models: In the light of the state of the art, there is no approach to learning high level abstraction time and concurrent task models (Planning Domain Description Language). Such a language is necessary to model the tasks of a cobot that can use 2 robotic arms like YuMi or in interaction with a human operator (representation of shared tasks).

Cobot acceptance and intelligibility: the Montreal declaration for a responsible AI, the question of ethics raised at the Innorobo exhibition or the «right to explanation« for algorithmic decisions which is now part of the law for a Digital Republic (2016) testify to the importance of this issue. One of our hypotheses is that the use of a cobot into the work environment must reconcile efficiency, performance, development of human skills and increase the psychological comfort of the operators: intelligibility of the cobot's decisions, sense of action control, control of attentional resources, ability of anticipation and reinforcement of the identity and social status of the operators and work collectives.

Grand, M.; Pellier, D.; Fiorino, H. TempAMLSI: Temporal Action Model Learning Based on STRIPS Translation. In Proceedings of the Thirty-Second International Conference on Automated Planning and Scheduling (ICAPS 2022), 2022.

Liang, Y.-S.; Pellier, D.; Fiorino, H.; Pesty, S. iRoPro: An interactive Robot Programming Framework. Int. J. Soc. Robot. 2021, 14 (7), 177–191.

Landry, A. Rendre les technologies émergentes favorables à l’activité : expériences d’accompagnement de la cobotique industrielle. Symposium presented at the 22e congrès de l’Association Internationale de Psychologie du Travail de Langue Française (AIPTLF), Montréal, Québec, 2023.

Cippelletti, E.; Fournier, E.; Landry, A. Acceptabilité des robots collaboratifs (Cobot) par des travailleurs français. In A. Landry: Rendre les technologies émergentes favorables à l’activité : expériences d’accompagnement de la cobotique industrielle [Symposium]. Symposium presented at the 22e congrès de l’Association Internationale de Psychologie du Travail de Langue Française (AIPTLF), Montréal, Québec, 2023.

The ambition of cobotics (or collaborative robotics) in industrial environments is to "get the industrial robot out of its protective cage" to work close to human operators on the same workstation. Man and Robot work safely together, side-by-side, on tasks combining physical effort and human expertise (e.g. stripping, assembling parts, packaging/palletizing parts that require taking into account the context and its hazards). These cobots are expected in particular by Small and Medium Industries (SMI) in order to meet their needs for flexibility. Indeed, SMIs are subject to the increasingly rapid renewal of their products and a growing demand for increasingly personalized products.

The general objective of the prog4Yu project is to develop a "programming by demonstrations" approach to cobots for production operators who are not experts in programming languages but have the expertise on performing production tasks, while ensuring that this approach is acceptable and intelligible to these operators. To successfully carry out the various facets of this multidisciplinary project, the consortium includes academic partners in computer/robotics and ergonomic psychology as well as an industrial partner in the field of robotics.

From an industrial point of view, the objective of the prog4Yu project is to develop an industrial cobot demonstrator "programmable by demonstration" to deal with the problem of packaging/palletizing parts at the end of the production line with the YuMi cobot.

From a scientific point of view, the objective is to (1) develop algorithms for learning cobot trajectories (gestures) and intelligible task models (semantics of gestures) according to a demonstration approach and (2) identify the factors that in this approach promote or hinder the acceptance of cobotics in an industrial environment. The operator demonstrations are done by kinesthetic manipulations of the cobot arms and via a multimodal Human-Robot IDE (Integrated Development Environment) interface.

The issues addressed by the project are:

1. Trajectory learning by demonstration: learning trajectories from few demonstrations (kinesthetic manipulation of a robotic arm by the operator), in an incremental and efficient way, i.e. by allowing a fluid interaction with the operator is an important scientific issue that we will have to solve to implement the project's industrial demonstrator.

2. Learning task models: In the light of the state of the art, there is no approach to learning high level abstraction time and concurrent task models (Planning Domain Description Language). Such a language is necessary to model the tasks of a cobot that can use 2 robotic arms like YuMi or in interaction with a human operator (representation of shared tasks).

3. Cobot acceptance and intelligibility: the Montreal declaration for a responsible AI, the question of ethics raised at the Innorobo exhibition or the "right to explanation" for algorithmic decisions which is now part of the law for a Digital Republic (2016) testify to the importance of this issue. One of our hypotheses is that the use of a cobot into the work environment must reconcile efficiency, performance, development of human skills and increase the psychological comfort of the operators: intelligibility of the cobot's decisions, sense of action control, control of attentional resources, ability of anticipation and reinforcement of the identity and social status of the operators and work collectives.