Skip to main content

​
Constellation DAG

Single-satellite planning assigns steps to orbital windows on one spacecraft. Constellation DAG planning distributes a workload across multiple satellites, routing intermediate data through inter-satellite links (ISL) and ground relays to minimize end-to-end latency.
When to use this — Any workload that benefits from parallelism across satellites, or where a single satellite lacks the resources (compute, power, storage, contact time) to complete the job within the deadline.

​
Architecture

Constellation planning is built on two core components:

​
ContactGraph

The ContactGraph builds a time-expanded graph of all communication opportunities across the fleet within the planning horizon.
CapabilityDetails
ISL detectionLine-of-sight between satellite pairs, accounting for Earth obstruction
Ground pass detectionVisibility windows to all 12 ground stations per satellite
Routing algorithmTime-expanded Dijkstra — finds shortest-latency path through ISL and ground hops
Fleet constructionbuildContactGraph generates the full fleet graph from TLE data
The graph edges are weighted by link capacity (data rate x duration) and propagation delay. Dijkstra finds the minimum-latency path for transferring data from any satellite to any other satellite or ground station.

​
ConstellationPlacer

The placer runs a 5-phase algorithm to assign steps to satellites:
PhaseDescription
1. PreparationPropagate orbits, build contact graph, compute per-satellite resource envelopes
2. Greedy placementScore each satellite for each step (resource fit, data locality, contact opportunity) and assign greedily by priority
3. ReplicationFor critical steps, place redundant copies on backup satellites for fault tolerance
4. Schedule assemblyBuild the final timeline with ISL transfer segments between cross-satellite dependencies
5. MetricsCompute critical path, reliability estimate, and resource utilization
Scoring considers four factors: available compute capacity, proximity to upstream data, upcoming contact windows, and current battery state-of-charge.
For fleets of 10 or more satellites, the placer uses a k-d tree spatial index to accelerate nearest-neighbor queries during ISL detection. This keeps planning time sub-linear with fleet size.

​
ISL Transfer Segments

When two dependent steps are placed on different satellites, the planner inserts ISL transfer segments. These are scheduled during mutual visibility windows and include FEC overhead. 
{
  "type": "isl_transfer",
  "from_satellite": "25544",
  "to_satellite": "48274",
  "data_mb": 36.75,
  "fec_overhead_mb": 1.84,
  "link_budget": {
    "distance_km": 1240,
    "data_rate_mbps": 10,
    "duration_s": 31
  },
  "window": {
    "start": "2026-03-10T08:12:00Z",
    "end": "2026-03-10T08:18:30Z"
  }
}
If no direct ISL path exists, the contact graph routes through intermediate satellites or ground relay (satellite A downlinks to a ground station, which uplinks to satellite B during a later pass).

​
API Usage

​
Create Constellation Plan

curl -X POST https://rotastellar-cae.subhadip-mitra.workers.dev/v1/constellation/plan \
  -H "Content-Type: application/json" \
  -H "Origin: https://rotastellar.com" \
  -d '{
    "satellite_ids": ["25544", "48274", "55909"],
    "custom_job": {
      "name": "Distributed EO Pipeline",
      "steps": [
        {
          "id": "capture_1",
          "name": "Capture Region A",
          "location": "onboard",
          "compute_s": 60,
          "output_mb": 2000,
          "depends_on": []
        },
        {
          "id": "capture_2",
          "name": "Capture Region B",
          "location": "onboard",
          "compute_s": 60,
          "output_mb": 2000,
          "depends_on": []
        },
        {
          "id": "merge",
          "name": "Merge and Mosaic",
          "location": "onboard",
          "compute_s": 180,
          "input_mb": 4000,
          "output_mb": 500,
          "depends_on": ["capture_1", "capture_2"]
        },
        {
          "id": "downlink",
          "name": "Deliver Results",
          "location": "ground",
          "depends_on": ["merge"]
        }
      ],
      "policy": { "objective": "min_latency", "deadline_orbits": 6 }
    }
  }'

​
Request Parameters

ParameterTypeRequiredDescription
satellite_idsstring[]YesNORAD catalog IDs of satellites in the constellation
preset_idstringNoUse a preset workload (mutually exclusive with custom_job)
custom_jobobjectNoCustom DAG definition
policy.objectivestringNomin_latency, balanced, max_reliability
policy.deadline_orbitsnumberNoMaximum orbits to complete the workflow
replication_factornumberNoNumber of redundant copies for critical steps (default: 1)

​
Response Structure

The response extends the standard plan format with constellation-specific fields: 
{
  "id": "const-7b2e4f...",
  "constellation": {
    "satellites_used": 3,
    "satellites_available": 3,
    "isl_transfers": 2,
    "ground_relays": 0
  },
  "critical_path": {
    "steps": ["capture_1", "isl_transfer_1", "merge", "downlink"],
    "duration_s": 4320,
    "bottleneck": "isl_transfer_1"
  },
  "metrics": {
    "total_compute_s": 300,
    "total_transfer_s": 186,
    "isl_data_mb": 2038.5,
    "ground_data_mb": 525,
    "reliability": 0.964,
    "satellites_used": 3
  },
  "placements": [
    { "step_id": "capture_1", "satellite_id": "25544", "reason": "best_resource_fit" },
    { "step_id": "capture_2", "satellite_id": "48274", "reason": "best_resource_fit" },
    { "step_id": "merge", "satellite_id": "55909", "reason": "data_locality" }
  ],
  "plan": { ... },
  "events": [ ... ]
}

​
Metrics

MetricDescription
critical_path.duration_sLongest sequential chain through the DAG including transfers
critical_path.bottleneckThe step or transfer that dominates the critical path
satellites_usedNumber of satellites with at least one assigned step
isl_transfersNumber of inter-satellite link transfers in the plan
reliabilityCombined delivery confidence across all paths

​
Console Integration

The Console displays constellation plans in a DAG tab powered by React Flow:
  • Each satellite is a swim lane
  • Steps are nodes, colored by satellite assignment
  • ISL transfers appear as animated edges between lanes
  • The critical path is highlighted
  • Clicking a node shows step detail, resource usage, and the scoring breakdown from the placement phase
Constellation planning requires valid TLE data for all satellites in the request. If any satellite ID is not found in the CelesTrak catalog, the request returns a 400 error with the missing IDs listed.

Pareto Planning

Multi-objective optimization for constellation plans

Agent Protocol

Agent constellation mode for multi-satellite execution