A deep reinforcement learning and a dynamic graph neural network-based scheduling agent to control a multi-task robot
• A scheduling problem representative of today’s robotic-based production environments and challenges. • A mathematical formulation for a optimization model. • A framework including a simulator of the robot’s movements and a custom solving strategy based on Q-values guided beam search. • A dynamic and heterogeneous graph representation and a custom graph neural network. • A dedicated training algorithm based on reinforcement learning ( ɛ ɛ -greedy DQN).
Authors
- Adnène Hajji (ORCID: https://orcid.org/0000-0002-1689-0939)
- Mohamed Farah (ORCID: https://orcid.org/0000-0003-0530-2086)
- Anas Neumann (ORCID: https://orcid.org/0000-0003-2872-2331)
- Monia Rekik (ORCID: https://orcid.org/0000-0002-8731-1965)
- Hedi Boukamcha
- Gabriel Caron Guillemette
Institutions
- Center for Interuniversity Research and Analysis on Organizations (CA)
- Decision Systems (United States) (US)
- Polytechnique Montréal (CA)
- Manouba University (TN)
Publication Details
- Journal
- Robotics and Computer-Integrated Manufacturing
- Published
- 2026-05-22
- DOI
- https://doi.org/10.1016/j.rcim.2026.103328
- Primary Topic
- Reinforcement Learning in Robotics
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Natural Sciences and Engineering Research Council of Canada