SWARMOS: Secure & Resilient Swarm Orchestration System

SWARMOS is a production-grade research framework designed for decentralized multi-agent coordination in high-stakes environments. It extends the Consensus-Based Bundle Algorithm (CBBA) with a Strategic-Grade Anomaly-Aware Filter to ensure mission continuity under extreme communication degradation, adversarial bid-poisoning, and kinetic attrition.

1. System Workflow & Architecture

SWARMOS operates on a layered defense-in-depth architecture. Every coordination message passes through multiple strictly-defined physical and strategic guardrails before influencing the collective fleet state.

graph TD
    subgraph "External World"
        A[Adversarial Agents] -->|Poisoned Bids| N[Degraded RF Network]
        E[Environment] -->|Obstacles/Threats| N
    end

    subgraph "SWARMOS Node Architecture"
        N -->|Telemetry| SC[Safety Compiler]
        SC -->|Numerical Validation| SAF[Strategic Anomaly Filter]
        SAF -->|Kinematic Heuristics| CBBA[Resilient CBBA Engine]
        CBBA -->|Dynamic Recovery| RM[Recovery Module]
    end

    subgraph "Outcomes"
        RM -->|Re-allocation| T[Task Completion]
        SAF -->|Trust Score Decay| Q[Quarantine & Isolation]
    end

    style SC fill:#f96,stroke:#333,stroke-width:2px
    style SAF fill:#f9f,stroke:#333,stroke-width:4px
    style CBBA fill:#bbf,stroke:#333,stroke-width:2px

Loading

2. Core Capabilities

🛡️ Strategic Resilience

• TypeScript → Python SafetyCompiler Bridge: Seamless integration connecting the TypeScript API gateway (server.ts) directly to the canonical Python SafetyCompiler subprocess as the single source of truth, enforcing fail-closed bounds checking on all LLM-decomposed mission manifests.
• Strategic-Grade Anomaly Filter: Heuristic-based detection that validates bids against physical hardware limits (Max Velocity, Path-Loss, Reward Bounds).
• Subsystem Telemetry (comms_transceiver): Granular multi-dimensional health diagnostics tracking RF transceiver attenuation (comms_transceiver), propulsion degradation, and GPS spoofing in real-time.
• Automated Quarantine: Nodes identified as anomalous are isolated from the consensus pool until they demonstrate consistent kinematic honesty (Remediation logic).
• Numerical Hardening: The Safety Compiler acts as a fail-closed firewall, rejecting all non-finite (NaN, Inf) or physically impossible payloads.

🔬 High-Rigor Research Engine

SWARMOS features a specialized empirical engine for high-confidence research:

• Statistical Significance: Custom implementation of Welch's T-Test and Cohen's d (Effect Size) to validate performance gains.
• Monte Carlo Sweeps: Supports 50+ seeds per configuration with 95% Confidence Interval (CI) reporting.
• Empirical 50% Attrition Level (loss_50_catastrophic): Rigorously models an exact 50% kinetic fleet loss combined with 50% stochastic RF packet drop and electronic warfare jamming bubbles.
• Failure Envelopes: Automated "Stress Searching" to identify the precise packet-loss thresholds where coordination breaks down.
• Ablation Infrastructure: Systematic toggling of modules to isolate the exact source of resilience.

3. High-Rigor Research Results (Artifact v2.1.0)

Our latest 1,400-trial full matrix benchmark sweep and 450-trial systematic ablation study highlight the resilience of the SWARMOS coordination architecture:

Scenario / Configuration	Algorithm	TCR (Mean ± CI)	Significance ($p$)	Effect Size ($d$)
Catastrophic Attrition (50% Fleet Loss + 50% RF Drop)	SWARMOS	98.0% ± 3.5%	0.0392 (*)	1.57 (Large)
Catastrophic Attrition (FS=4, T=10)	Static Baseline	74.0% ± 4.3%	0.0481 (*)	-1.41
Adversarial Injections (Poisoned Bids)	SWARMOS	99.3%	p < 0.05	Multi-tier Recovery
High Interference Breakdown Threshold	SWARMOS	Stable up to 70% Loss	—	Degrades only at 80%

Component Ablation Breakdown

• Dynamic Recovery Module: Contributes +2.0% TCR baseline gain under severe fleet attrition.
• Strategic Anomaly Filter: Quarantines poisoned bids and adversarial nodes, ensuring 99.3% TCR under targeted sabotage.
• Safety Compiler: Zero-tolerance deterministic bounds rejection protecting the consensus engine from malformed/out-of-bound missions.

Full multi-config tables, breakdown threshold sweeps, and paired effect sizes are available in docs/RESEARCH_REPORT_RIGOR.md.

4. Getting Started

Directory Structure

• swarmos/swarm_engine/: Physics and Resilient CBBA core.
• swarmos/ai_layer/: Safety Compiler and Anomaly Filter.
• swarmos/scripts/: High-rigor benchmark runners and report generators.
• swarmos/utils/: Statistical analysis library and loggers.
• docs/: Rigorous documentation and metrics definitions.

Running the Benchmark Suite

To reproduce the high-rigor statistical report:

PYTHONPATH=. python3 swarmos/scripts/run_rigorous_bench.py

To perform a systematic ablation study:

PYTHONPATH=. python3 swarmos/scripts/ablation_study.py

5. Citations

• Choi, H. L., et al. (2009). "Consensus-based decentralized auctions for robust task allocation." IEEE Transactions on Robotics.
• SWARMOS Research Group. (2026). "Statistical Resilience in Decentralized Swarm Coordination."

6. Ecosystem Alignment

SWARMOS is built with cross-platform scalability in mind, aligning with the highest standards of our partner ecosystems:

• NVIDIA Developer: Architected for future CUDA-accelerated kinematic validation.
• AWS Builder: Optimized for massively parallel Monte Carlo simulation.
• Google Developer: Integrated with Gemini-powered strategic mission logic.

---

Developed by a First-Year IIT Madras Student Researcher. Protected by SWARMOS Academic License.

⚖️ Intellectual Property & Legal Protection

SWARMOS is the intellectual property of Shivam Singh (IIT Madras).

• Restricted Use: This framework is released under a custom Research-Only License. Commercial use, redistribution, or unauthorized derivation is strictly prohibited.
• Anti-Plagiarism: Any attempt to copy or claim this work as your own will be met with legal and academic action.
• Citation Required: Any research leveraging this code must cite:

  Singh, S. (2026). SWARMOS: Secure & Resilient Swarm Orchestration System. IIT Madras.

For commercial licensing inquiries, please contact the author.