With the rapid advancement of artificial intelligence, robots are being deployed across more domains than ever before. However, most industrial robots remain limited to repetitive tasks in controlled settings. They struggle with adaptive decision-making in the complex, unstructured real world because of a fundamental flaw: the von Neumann architecture. This traditional system separates memory and computation, leading to inefficient data shuttling and a "memory wall" that hinders real-time processing. To address this, researchers from Shandong University are championing a shift toward neuromorphic computing. By mimicking the event-driven processing of the human brain, these new systems integrate memory and computation, enabling robots to process information in parallel with minimal energy use.
The Hardware of "Thinking" Machines
The review explores various material systems, including metal oxides, low-dimensional materials, and organic polymers, that are used to build artificial synapses and neurons. These devices, such as memristors and transistors, can retain information and adjust their "synaptic weights" just like biological systems.
1.Metal Oxides: Favored for their multi-modal sensing and nonvolatile memory, ideal for complex motion control.
2. Low-Dimensional Materials: Offer high integration density and low energy consumption for fast computing.
3.Organic Materials: Provide excellent biocompatibility and flexibility, making them perfect for "electronic skins" and wearable robotics.
From Perception to Action
The research classifies the application of these devices into three levels of Embodied Intelligence:
Low-Level: Focused on basic environmental perception, such as real-time tactile processing in artificial skin.
Mid-Level: Enabling human-robot interaction through gesture recognition and feedback systems.
High-Level: Achieving fully autonomous, closed-loop control, including navigation in complex mazes and millisecond-latency flight control for drones.
The Path Forward
While the potential is transformative, challenges remain in large-scale integration and device reliability. "The ultimate objective," the authors note, "is to realize neuromorphic systems with lifelong learning abilities and advanced cognitive functions". Within the next five years, the focus will shift toward software-hardware co-design, allowing robots to navigate dynamic environments with situational awareness comparable to biological organisms.
SmartBot
Systematic review
Brain-Inspired Neuromorphic Device for Artificial Intelligent Robots Applications
13-May-2026