Metabolic Parsimony as a Consequence of Optimal Viability Control

SUMMARY: This study presents a controlled computational experiment comparing the Biological Principle of Least Action (BPLA) optimal control law against a standard Q-learning (RL) agent. The experiment was conducted in a stochastic 2D environment with explicit metabolic costs, Gaussian observation noise, and an extinction boundary. CORE FINDINGS: The results provide strong empirical evidence for the "metabolic parsimony" predicted by the BPLA framework. 15-FOLD EFFICIENCY GAIN: The BPLA agent expended approximately 15-fold less total action cost compared to the Q-learning agent (2. 8 +/- 0. 9 vs. 42. 0 +/- 25. 3 units). STATISTICAL SIGNIFICANCE: The efficiency gap is highly significant (Mann-Whitney U = 146, p = 3. 2 x 10^-10, Cohen's d = 2. 19). SUPERIOR MASS ACCUMULATION: BPLA agents accumulated significantly more replicative mass by episode termination (median 1, 205 +/- 738 vs. 644 +/- 552; p = 0. 0012). CONSTRAINED PERSISTENCE: While binary survival rates were similar (87. 5% vs 80. 0%), BPLA demonstrated a unique ability to scale motor effort inversely with existential risk (the "Urgency" term), a feature missing from standard RL architectures. CONCLUSION: incorporating explicit biophysical viability constraints into the objective functional of autonomous agents offers massive efficiency gains in resource-limited settings. These results suggest that adaptive agency is not merely reward maximization, but the minimal admissible state-space flow required to maintain physical persistence. TECHNICAL SPECIFICATIONS: Environment: 50x50 continuous 2D arena with stochastic resource drift. Metabolism: Basal decay (0. 15/step) + per-unit action cost (0. 08/unit). Trial Count: n=40 independent trials per condition (600 timesteps each). Reproducibility: All trials used a fixed pseudo-random seed (42). KEYWORDS: Optimal Control, Metabolic Cost, Reinforcement Learning, Viability Theory, Adaptive Behaviour, BPLA, Computational Biology. DATA AND CODE AVAILABILITY: All simulation code and raw data are timestamped and available for verification at the repository link provided here.