Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back Dynamic Graph Reinforcement Learning for Joint Task Assignment and Joint Trajectory Planning in Space Debris Removal by Multiple Spacecraft DANDAN SU, K.A. NEUSYPIN, DUN GE An innovative end-to-end framework integrating dynamic graph representations and multi-agent reinforcement learning is proposed for cooperatively solving target allocation and trajectory planning problems. A topologically adaptive graph neural network is developed that explicitly incorporates the physical constraints of orbital mechanics into the message aggregation architecture. An algorithm with cooperatively-oriented advantage estimation is synthesized within the paradigm of centralized learning with decentralized execution. A dynamic graph representation of state is created that supports emergent agent behavior. Verification on real-world data from IRIDIUM-33 debris confirmed the qualitative superiority of the proposed approach: an increase in the clearance rate and a reduction in fuel consumption were achieved while maintaining subsecond decisionmaking time, demonstrating the potential for onboard implementation. Keywords: spacecraft, space debris; graph neural networks; cooperative trajectory planning. DOI: 10.25791/pribor.12.2025.1639 Pp. 39-50.