MH-FLOCKE: Biologically Grounded Embodied Cognition Through a 15-Step Closed-Loop Architecture for Quadruped Locomotion Learning

Abstract

We present MH-FLOCKE, an embodied AI platform in which simulated quadruped creatures learn locomotion through a biologically grounded cognitive architecture. Unlike end-to-end reinforcement learning approaches that treat the body as an optimization target, MH-FLOCKE implements a 15-step closed-loop processing cycle that integrates proprioception, embodied emotions, episodic memory, motivational drives, a Global Workspace for attentional competition, metacognitive self-assessment, and reward-modulated spike-timing-dependent plasticity in a spiking neural network. Systematic ablation across 60+ runs isolates the contribution of each component: vestibular reflexes eliminate all falls (27 falls to 0), motor babbling increases flat-terrain distance by 763%, a cerebellar forward model produces measurable prediction errors for the first time, and an olfactory sensory environment enables stimulus-driven behavior switching (22 scent sources found vs. 0 without sensory input). We report an interaction effect where olfactory steering interferes with cerebellar learning, reducing full-system performance by 10%, while the same sensory system rescues runs without a cerebellum by 277%. The architecture is implemented as an open-source framework building on Integrity-OS, with MuJoCo physics and a custom binary logging format (FLOG) for reproducible analysis. All ablation data are publicly available. Changes (March 2026): Added PPO baseline comparison (Table 3), multi-seed validation (3 seeds, 24 runs), mathematical formulations (Section 3.9), cross-embodiment transfer to Unitree Go2 (Section 5.6), and companion video. Softened novelty claims per reviewer feedback. New sections: 5.6, 5.7, 6.5, 6.6. Contact updated to [email protected].

Keywords