Flexible, super-resolution skin with in-sensor hyperdimensional computing for real-time pressure field perception

Precise, highly spatial resolution surface pressure measurement is critical for reliable state assessment of unmanned systems, such as the flight safety of unmanned aerial vehicles. Conventionally, the acquisition of such high-resolution data requires dense, expensive sensing arrays or relies on external computational resources. Here, we introduce a flexible electronic skin featuring a sparse sensing network integrated with in-sensor numerical hyperdimensional computing, thereby facilitating super-resolution sensing and in-sensor learning. This innovative approach processes sparse inputs to produce high-resolution pressure field maps on-device, enabling rapid real-time analysis and precise flight state identification. The numerical hyperdimensional computing enhances the efficiency of in-sensor learning, overcoming the limitations of conventional vector symbolic architectures previously confined to classification tasks. Furthermore, the integration of hyperdimensional computing substantially boosts the performance of the flexible sensing skin. Compared to the computer-aided super-resolution method, our approach reduces power consumption from ~40 watts to ~0.09 milliwatts, memory usage from ~2274 to ~32 kilobytes, and latency from ~115 to ~10 milliseconds, respectively. The achieved super-resolution capability reaches an enhancement factor of 5.517. Compared with conventional high-density sensing arrays, our approach reduces the number of interconnects by 91.5% (using a 4 by 4 array). Last, the developed super-resolution skin is applied in unmanned aerial vehicle flight control.