Homeostatic Event-Driven Control Architecture: Autonomous Central-Place Foraging, Binary Memory Indexing, and Swarm Federation

Abstract

Autonomous robotic exploration in resource-constrained, unstructured environments is often bottlenecked by the high computational overhead, sample inefficiency, and opacity of deep reinforcement learning (RL). This paper presents the Homeostatic Event-Driven Control Architecture (HEDCA)—a transparent, compute-efficient paradigm grounded in cybernetics, biological homeostasis, and case-based episodic memory reuse. HEDCA monitors physiological vitals via ternary event-driven neurons with adaptive noise-floor deadbands, projecting state deviations into a sparse Meaning Space to generate compact, binary-quantized Relationship Vectors (RVbin) requiring only 256 bytes per episode—a memory reduction of >99.9% compared to raw continuous trajectories. Closed-loop error nullification is achieved via an energy-aware Experience Map evaluated via bitwise Hamming distance, accelerating query times to ≤10.3 μs. The system introduces a Dynamic Energy Return Horizon (Leash) and a Closed-Loop Heading Gate to prevent spatial divergence. Key extensions include:- Nested Spike Handling (fallback actions)- Memory Temperature (usage-based heat)- Consolidation (hot-sharpening, cold-compression)- Meaning Nodes (sensor-correlation detectors that influence global caution) Across sequential empirical benchmarks (395,000 simulation steps) and swarm scaling tests (N=1 to 16), HEDCA achieves 100% survival rates, near-linear collective knowledge accumulation (O(N^1.09)), +92.3% exploratory play gain, and a total RAM footprint under 35 KB, operating entirely without offline pre-training or external reward engineering. A visual tool, FossilView, renders the memory tree, revealing the system's internal structure.