在地震、山崩或大火等灾难救援中，超小型无人机能进入倒塌建筑的缝隙寻找生还者，但因其重量轻且运算与电力有限，在浓烟、夜间或面对透明玻璃时，传统相机与光达感测器往往会失效。为了解决这个问题，荷兰代尔夫特理工大学的研究团队从老鼠与猫等夜行性动物的胡须获得灵感，开发出一种轻量、低延迟且低功耗的触觉感测系统，让无人机即使在失去视觉的情况下也能顺利导航。

这款触觉感测器的「胡须」由直径仅0.4毫米的镍钛合金细线制成，研究人员在巴掌大的无人机上安装了两根长约8英寸的胡须，并在根部配置三个小型压力感测器，透过侦测细线弯曲程度并由机载微控制器即时计算与障碍物的距离和位置。团队经过72次飞行测试后，发现将胡须以45度角安装最能维持飞行稳定并兼顾前方探测范围；此外，针对螺旋桨震动杂讯导致高达38%误判率以及感测器数值随时间偏移（漂移）的问题，团队开发了能即时估计漂移趋势的校正演算法，结合滤波技术后成功将误报率降至0%。

在排列著透明墙壁的测试环境中，无人机能够在触碰墙面后保持60至100毫米的距离平行飞行，甚至能应对倾斜墙面，这是雷射感测器无法做到的。为了避免在真实灾难现场的复杂瓦砾或房间角落中被困住，研究人员进一步设计了能以机率估计墙壁形状并优先探索未知区域的演算法，使无人机在玻璃房间中成功找到约76公分宽出口的机率达到90%，且整套触觉感知系统仅占用机载192 KB记忆体中的34 KB，展现了极高的扩充与应用潜力。

In disaster rescue operations such as earthquakes, landslides, or major fires, ultra-compact drones can fit into the crevices of collapsed structures to locate survivors, but their limited compute and power capacities cause traditional cameras and lidar to fail in smoke, darkness, or when facing clear glass. To overcome this limitation, a research team at Delft University of Technology in the Netherlands drew inspiration from the whiskers of nocturnal animals like mice and cats to develop a lightweight, low-latency, and low-power tactile sensor system that enables drones to navigate even when they cannot see.

The drone's artificial whiskers are crafted from delicate nickel-titanium wires measuring 0.4 millimeters in diameter, with two nearly 8-inch-long whiskers attached to a palm-sized drone and equipped with three small pressure sensors at the base to assess wire bending and calculate barrier distance and position via an onboard microcontroller. After comparing 72 flights across various angles and speeds, the researchers adopted a 45-degree mounting angle to minimize orientation deviation while preserving forward sensing range; furthermore, to solve the 38 percent false-positive rate caused by propeller vibrations and sensor drift over time, they developed an in-flight correction algorithm that estimates drift trends and, when combined with filtering, reduced false positives to 0 percent.

In tests featuring transparent walls, the drone successfully flew parallel to the surfaces at a distance of 60 to 100 millimeters upon contact and navigated angled walls that laser sensors would miss. To prevent the drone from getting stuck in corners or complex real-world rubble, the team encoded a probabilistic exploration method that estimates wall shapes and prioritizes unexplored areas, allowing the drone to find a 2.5-foot-wide exit in a glass-walled room in 90 percent of tests while using only 34 KB of its 192 KB onboard memory for real-time tactile perception.