Aerospace Instrument-Making Annotation << Back Graph-Based Learning for Joint Planning of Active Space Debris Removal Missions Dandan Su, Neusypin K.A., Jiaman Ma The increasing debris density of low-Earth orbit (LEO) poses significant risks to the safety of space activities and the effectiveness of existing missions. Coordination of multi-vehicle active debris removal is complicated by the spatiotemporal uncertainty of trajectories and capture windows, which hinders the construction of physically feasible and optimal task graphs, and the combinatorial complexity of joint task assignment and routing as the number of debris increases. To overcome these limitations, we propose STAR-Graph – a spatiotemporal assignment framework based on a reachability graph that accounts for uncertainty. First, a dynamic reachability graph is constructed, where pairwise feasibility between targets is extracted through contrastive self-learning on orbital propagation and time windows. Then, differentiable deep soft clustering is applied to form schedulable task groups based on temporal reachability and spatial proximity. Finally, a cluster-driven adaptive wide-neighborhood search (ALNS)-based planner jointly optimizes assignments and routes, increasing the probability of successful execution and reducing the total maneuvering cost (ΔV). Experiments on real and simulated orbital scenarios demonstrate the superiority of STAR-Graph over basic heuristic and optimization methods in terms of fuel efficiency, completeness, and robustness. Keywords: spacecraft, space debris, STAR graph, mission effectiveness. DOI: 10.25791/aviakosmos.2.2026.1535 Pp. 13-24.