MSBAI (Microsurgeonbot Inc.) has secured a Direct-to-Phase II SBIR award—$1.2M over 18 months—to accelerate OrbitGuard, a hybrid-intelligence copilot for real-time Space Domain Awareness (SDA). The opportunity originated with the DoD Chief Digital and Artificial Intelligence Office (CDAO), was competitively selected by OSD, and is executed by the Air Force Digital Transformation Office (DTO). The effort positions OrbitGuard for operational use amid a swelling orbital population projected to surpass 17,000 active satellites by 2026.
Key Points
Funding & scope: $1.2M, 18 months, Direct-to-Phase II SBIR.
Mission: Real-time detection of anomalous satellite behavior and maneuver identification, integrated with DoD data services like UDL and running on MSBAI’s GURU platform.
Performance to date: Peer-reviewed evaluations cite 94–98% anomaly-detection/classification accuracy across ~15k on-orbit objects.
Roadmap under the contract: Scale to >20k RSOs, ~2-minute end-to-end latency, ≥99.9% uptime, and model upgrades (Graph-JEPA, Patch Time-Series Transformers, multi-agent RL).
Explainability & ATO: Blackboard-style logging, symbolic constraint checks, and NIST SP 800-53 artifacts to support Authority-to-Operate.
What OrbitGuard Is—and Isn’t
OrbitGuard is MSBAI’s neuro-symbolic decision aid for SDA. It fuses symbolic rule-based checks with learned world models and reinforcement-learning (RL) planners to move operators from alert → investigation → visualization within seconds, emphasizing traceability over prompt-based heuristics. Data sources include infrared, electro-optical feeds, and CelesTrak/Space-Track catalogs.
Naming note: Do not confuse MSBAI’s OrbitGuard software with Infinite Orbits’ “Orbit Guard” (a GEO-capable inspection/SSA smallsat). They are unrelated products that share a similar name.
Why Hybrid-Intelligence Matters for SDA
Neuro-symbolic approaches combine deterministic logic (for hard constraints and auditability) with self-supervised world models (JEPA family) that learn the dynamics of orbital behavior without requiring labeled data. JEPA (Joint-Embedding Predictive Architecture) predicts representations of missing context rather than pixels/tokens, improving robustness and reducing hallucinations—attractive traits for high-stakes ops.
MSBAI’s roadmap explicitly calls out Graph-JEPA (to capture relational structure among satellites) and Patch Time-Series Transformers (for time-localized anomaly signals), coupled with multi-agent RL for adaptive courses of action.
Data Fabric & DoD Integration
OrbitGuard is designed to live where the SDA enterprise lives: the U.S. Space Force’s Unified Data Library (UDL)—a cloud repository that standardizes schemas, aggregates commercial/government data, operates across classification levels, and connects directly to sensors like Space Fence.
UDL capabilities (per SSC): manages commercial SDA data enterprise-wide; agnostic to visualization/analytics; controls and delivers data across multiple enclaves; Authority-to-Operate at U/S/TS.
UDL’s ecosystem includes feeds derived from Space-Track/CelesTrak (TLE/GP data) commonly used for custody and conjunction screening.
MSBAI’s PR also highlights a live demo at Space Systems Command’s SDA TAP Lab Apollo Accelerator (Colorado Springs) and autonomous integration with KeepTrack for tracking/visualization—aligning with TAP Lab’s mission to fast-track space battle-management software.
Performance, Roadmap & Operator Workflow
Current results: 94–98% detection/classification accuracy over ~15k objects; near-real-time detection of anomalous behaviors and precise maneuver identification.
Planned upgrades: Scale to >20k RSOs, ≈2-minute E2E latency, ≥99.9% uptime; Graph-JEPA + Patch TST improved predictors; explainability through blackboard logs and symbolic audits; ATO-ready prototype with NIST SP 800-53 controls.
Workflow: Trigger → fused evidence package (symbolic checks + JEPA world-model deviations + RL-suggested actions) → link-out to visualization (e.g., KeepTrack) → operator sign-off.
Competitive Landscape & Differentiators
OrbitGuard focuses on decision intelligence and explainable autonomy rather than owning a proprietary sensor network.
LeoLabs operates a proliferated phased-array radar network for persistent LEO tracking (fixed and expeditionary radars), providing low-latency orbital data-as-a-service and sovereign radar options.
ExoAnalytic Solutions fields the world’s largest commercial optical telescope network (EGTN) with 275–350+ telescopes for GEO/HEO/MEO surveillance and brightness/angles measurements.
COMSPOC provides high-fidelity SSA analytics (e.g., SSASuite with EKF smoothing, enterprise catalog ops).
NorthStar is deploying space-based optical SSA (“Skylark”) to observe space from space, aiming for continuous custody across regimes.
Differentiator: MSBAI’s hybrid-intelligence layer is sensor-agnostic, natively integrated into UDL, and oriented to explainable, auditable decision support—key for ATO and mission acceptance.
Context: The U.S. SDA Backbone
The Space Surveillance Network (SSN) blends radar/EO/RF assets; the Space Fence S-band radar (Kwajalein) provides uncued, highly sensitive detection and feeds SDA pipelines that UDL can expose across enclaves. OrbitGuard’s role is to interpret this deluge, flag outliers, and recommend actions, not to replace primary sensors.
Risks, Limitations & What to Watch
Data dependence & latency: Sensor availability and ingestion latency still bound detection timelines; integration across enclaves can be non-trivial. (UDL mitigates this with standardized schemas and cross-domain delivery.)
Ops trust: Even with blackboard logs and symbolic audits, operators must calibrate trust in JEPA/RL elements—an ongoing human-on-the-loop challenge.
Naming collision: Potential confusion with Infinite Orbits’ Orbit Guard smallsat underscores the need for clear branding in joint ops centers.
Outlook
MSBAI’s push aligns with the Space Force’s accelerator culture (SDA TAP Lab) and the broader shift toward AI-assisted space battle management. If OrbitGuard delivers its latency, uptime, and explainability targets, it could become a decision-grade copilot alongside incumbent sensor/analytics providers—particularly attractive for operations centers accessing data via UDL.
