Large-scale iontronic skins with neural-hierarchical processing architecture

The human skin can encode and gather signals from over 200,000 afferents, providing high spatiotemporal-resolution sensations to enable fine touch and to prevent skin injury. Existing robotic skins typically employ a serial architecture to readout multichannel signals, while such an architecture becomes incapable when a skin has thousands of receptors, resulting in readout latency, signal crosstalk, and bandwidth consumption. Here, we introduce a large-scale neural-hierarchical iontronic skin (NI-skin) with >10,000 sensors that mimics the hierarchical processing architecture of the tactile nervous system, transmitting tactile information in layers via signal encoding, convergence, and cognitive modeling. The position-encoding of the frequency-dependent iontronic receptors enable parallel transmission with ultralow latency and suppressed crosstalk, while the backend convergent compression minimizes bandwidth, achieving an 800 µm spatial resolution and a 10.6 ms temporal resolution. We demonstrate that robots equipped with our system can perform challenging tasks, including fine tactile interactions and the perception of pain-related mechanical stimuli for proactive injury avoidance. This technology provides a large-scale tactile system with high spatiotemporal resolution, paving the way for humanoid robots to achieve human-level perception and manipulation.