Abstract

Brain-inspired computing is becoming increasingly relevant to robotics because autonomous machines must operate under strict limits of power, latency, memory and uncertainty. This opinion-style mini review argues that neuromorphic computing, event-driven sensing and spiking neural networks offer a principled route toward embodied intelligence that is more compatible with real-world robotic deployment than purely cloud-centered or over-parameterized approaches. We summarize why classically trained deep models still struggle in low-power, low-latency settings, discuss how neuromorphic methods can improve perception-action loops and outline the remaining barriers, including training stability, benchmarking, hardware-software co-design and safety validation. We contend that the most promising near-term path is not to replace main-stream artificial intelligence, but to integrate brain-inspired modules into robotic systems that require continuous sensing, adaptive control and energy-efficient autonomy.

Keywords:Neuromorphic computing; Embodied intelligence; Event-based sensing; Edge AI