juin 2018 :

Rien pour ce mois

mai 2018 | juillet 2018

Learning by Demonstration of Action Models for Cobot Programming

Accueil > Communauté GOSPI


The research project CIRCULAR is offering PhD thesis on :
- Learning by Demonstration of Action Models for Cobot Programming at Grenoble Alpes university, LIG Laboratory

Since their emergence, industrial robots are traditionally programmed off-line by experts in robotics using computerbased coding interfaces. Robots are then placed in working plants to be used by human operators in order to execute repetitive tasks complying with code instructions. When the tasks assigned to the robot have to change, this robot needs to be completely recoded by the robotics experts, which is
challenging, takes a lot of human efforts and time, and does not meet nowadays industry needs in supplychain reconfigurations and flexibility.
_To address this challenge, the PhD. Student recruited will have to devise a new human-friendly approach for industrial robot programming through on-site Human-Robot Interactions. The approach consists of three processes : human teaching,
robot learning and robot execution (see figure 1) : 1. The human operator that will use the robot teaches it a task by providing verbal instructions and by manipulating the robot’s effectors. 2. The robot learning process is based on learning by demonstration techniques [1]. In this process, the human operator and the robot build a common symbolic representation of the task : this symbolic representation is a planning domain description language1. This language uses preconditions and effects to describe changes in the robot’s context. 3. The robot task execution is controlled by human verbal instructions that express the objectives of the task that the robot has to achieve. Then, the robot automatically computes a sequence of gestures achieving these objectives : the human operator does not provide the “recipe”, i.e., a sequence of gestures to achieve the objectives ; the “recipe” is computed by Automated Planning techniques [2]. For this project, we will use the PDDL4J library (pddl4j.imag.fr). The development will be implemented on the Baxter robotic platforms."

Application deadline : 06/07/2018 at 17:00 (CET)

For further information about critera eligibility please download the file.

Learning by Demonstration of Action Models for Cobot Programming