Research ยท Papers
Papers
Preprints and reports on the systems behind orbital compute. We publish the work that makes a feasible plan, a surviving execution, and a distributed orbital run possible.
Published
When Is Availability-Aware Training Worth It? A Benchmark and Empirical Study of Interruption-Resilient Optimization Under Predictable Compute Schedules
OrbitTrace, a benchmark of 50 physics-grounded compute-availability traces from SGP4 propagation, and an empirical study showing that when optimizer state is preserved across a gap, competent checkpoint-and-resume matches uninterrupted training. Availability-aware optimizers pay off in only one narrow regime.
Constraint-Aware Execution Planning for Hybrid Space-Ground Compute Workloads
A planning system that produces feasible execution plans for hybrid space-ground compute workloads under power, thermal, communication, and compute constraints. Deployed in production as the CAE API.
Collaborate.
We work with researchers on orbital scheduling, on-satellite ML, and distributed training across the space-ground boundary.