Biohybrid robots, which integrate living biological components such as muscle or neural tissues with artificial mechanical structures, represent a distinct class of systems capable of highly adaptive and biologically inspired movements. Compared to conventional robots, these platforms harness the intrinsic properties of living tissues, including self-repair, high power-to-weight ratios, and natural responsiveness to biochemical and physical stimuli. However, selecting optimal control strategies involves navigating significant trade-offs between stability, learning efficiency, and implementation complexity. Biological actuators are fundamentally constrained by nonlinear dynamics and "physiological drift" caused by metabolic fluctuations and fatigue, necessitating controllers that can adapt in real-time to maintain functionality. This paper provides a systematic comparison of AI methods applied to biohybrid control over the past five years, including deep learning, reinforcement learning, hybrid intelligent control, and data-driven adaptive models. The study reveals that hybrid intelligent control currently offers the most practical and balanced solution by embedding AI adaptability within classical stability frameworks. By partitioning the controller into model-based and learning-based components, this paradigm maintains formal safety guarantees while exploiting AI for dynamic compensation. Nevertheless, its implementation remains constrained by high architectural complexity and the difficulty of formally validating the interaction between discrete AI logic and continuous biological feedback. Finally, future research directions such as metabolic-aware control and sustainable intelligent systems are discussed to provide theoretical guidance for advancing robust AI-driven biohybrid robotics.
61 (Views) 
12 (Downloads)