NASA Space to Soil | Sujjay Karthikeyan

The Challenge.

The NASA Space to Soil Challenge asks: how do you turn orbital remote sensing data into actionable, ground-truth soil health intelligence at field scale? Satellite imagery tells you where conditions look anomalous. It doesn't tell you what the soil actually contains — or why.

The concept closes that gap by coupling an adaptive SmallSat tasking layer with a coordinated ground drone swarm, using edge AI to bridge the two domains in near-real-time.

Orbital Layer

Adaptive SmallSat

Onboard AI

Edge Inference

Ground Layer

Drone Swarm

Sensing

Multispectral + In-Situ

Status

Phase 1 Submission

System Architecture.

Space-to-Ground Sensing Loop

Orbital

→

SmallSat Multispectral Imaging

NDVI, moisture index, thermal anomaly detection

Edge AI

→

Onboard Anomaly Segmentation

Priority zone extraction before downlink — reduces bandwidth 10×

Tasking

→

Swarm Dispatch Commands

Coordinates assigned to ground drones for in-situ sampling missions

Ground

→

Drone Swarm In-Situ Sampling

Multi-UAV coverage of flagged zones — soil sensor payloads per drone

Fusion

→

Orbital + Ground Truth Integration

Ground sample data closes the loop back to satellite model calibration

Key Design Decisions.

Onboard Edge AI Tasking Rather than downlinking full imagery for ground processing, the SmallSat runs lightweight segmentation onboard to identify priority zones before transmission. This compresses the decision latency from hours to minutes and cuts required downlink bandwidth drastically.

Adaptive Revisit Scheduling The satellite's imaging cadence self-adjusts based on detected anomaly density — high-priority agricultural zones get more frequent passes during critical growth windows without consuming fixed revisit budget on stable areas.

Swarm-as-Sensor-Network Ground drones aren't sent to collect a single sample — they operate as a distributed sensor network covering the flagged zone simultaneously, with each drone carrying a different sensor payload (NPK, moisture, pH) to maximize information density per mission.

Closing the Calibration Loop Ground truth samples flow back up to calibrate the satellite's spectral models — making the system progressively more accurate as it accumulates local data. The orbital and ground layers aren't just co-located; they improve each other over time.

Why This Ties Together.

This concept directly connects two threads running through the rest of my work: the multi-UAV coordination research at SmartNet Lab and the aerospace structure work on the VTOL and glider projects. A soil-monitoring drone swarm tasked by satellite is the same multi-agent coverage problem — with an orbital planner replacing a central ground station.

Convergence Point.

Space to Soil is where swarm intelligence meets Earth observation. The SmartNet Lab coverage optimization work, the VTOL UAV design experience, and the reinforcement learning research all point at exactly this kind of system. Phase 1 is the concept. The architecture is buildable with existing components.