Sorcerer: Designing Human-Machine Teaming for Lethal Autonomous Systems

DoD Directive 3000.09 requires meaningful human control over autonomous weapons systems. Most implementations treat this as a compliance checkbox — a software flag that can be overridden, a UI confirmation that can be bypassed, a policy constraint that exists in documentation but not in silicon.

Sorcerer was designed to make human control physically undefeatable.

Three-Tier Control Hierarchy

The Sorcerer AI control system operates on a three-tier hierarchy:

Tier 1 handles formation patterns — the pre-programmed behavioral library that governs routine autonomous operations. AUV patrol patterns, UAV surveillance configurations, combined surface/subsurface formations, high-sea adapted patterns. This tier operates continuously without human input.

Tier 2 is edge AI threat classification. 8 TOPS minimum processing with multi-modal sensor fusion and confidence scoring. When Tier 2 identifies a threat above confidence threshold, it can authorize autonomous target bracketing — repositioning two drones to opposing flanks of the target. Tier 2 authority stops there. It cannot authorize kinetic engagement.

Tier 3 is hardware override. A physical relay on the command bus. Not a software flag. Not a firmware setting. A physical relay that cannot be defeated by any software state, any network command, or any AI decision.

Why Hardware Override Matters

Software-based safety systems can be compromised. A sufficiently sophisticated adversary with access to the system's network, firmware, or AI model can potentially override software constraints. This is not a theoretical concern — it is the basis of an entire category of adversarial AI attacks.

A physical relay on the command bus cannot be compromised by software. It cannot be overridden by a network command. It cannot be bypassed by an AI model that has been manipulated or has reached an unexpected state. The physics of the relay are the safety guarantee.

Population-Density Engagement Mode

One of the most significant innovations in Sorcerer is the population-density engagement mode — the first system of its kind to implement this capability.

Before any Spear Water authorization, Sorcerer queries the sensor mesh for population density below the computed intercept point. If density exceeds threshold, Spear authorization is withheld regardless of threat confidence level. The system will not authorize kinetic engagement over populated areas.

This is not a policy constraint. It is an architectural constraint implemented at the AI decision layer, with the physical relay as the final backstop.

Third-Party Integration Without Command Authority

Sorcerer exposes a third-party integration API supporting REST/JSON, DDS, and STANAG 4586 — the NATO standard for unmanned system control. This enables integration with existing C2 infrastructure, naval combat management systems, and allied force networks.

The critical design constraint: the API provides recommendations only. It never commands effectors. External systems can provide targeting data, threat assessments, and engagement recommendations. The decision to authorize kinetic engagement remains within the Sorcerer hierarchy, subject to the physical override relay.

Vessel IMU Predictive Compensation

Operating autonomous drones from a moving vessel in rough seas introduces a fundamental control problem: the launch and recovery platform is itself in motion. Sorcerer addresses this through Extended Kalman Filter fusion of vessel IMU data with 0.5–1.0 second predictive pre-compensation.

The result is a 60–70% reduction in vessel-motion-induced positioning error at Sea State 5 conditions. The system anticipates where the vessel will be, not just where it is.

Patent Pending. SETEC Astronomy LLC. Travis Martin, Inventor.