Touch pad based on triboelectric nanogenerator are attracting attention in the field of wearable electronics and human-machine interaction. However, it is still a challenge to realize touch pads combining large-area metamorphic stretchability, high power triboelectric sensing, intelligent free-sliding recognition and availability in extreme environments. Here, we fabricate an ultra-stretchable triboelectric touch pad by integrating liquid metal and high performance hydrogel-based triboelectric sensor arrays, which achieves Transformer-assisted gesture recognition for drone flight direction control. The internal hydrogel electrode has great stretchability, electrical conductivity and anti-freeze properties. Encapsulated in Ecoflex with sandpaper micro-surfaces, it endows high triboelectric power density and pressure sensitivity. With screen-printed liquid metal, the integrated touch pad is fully ultra-stretchable and large-area strain stability. Combining Internet of Things, we build a Transformer-assisted human machine interface. Notably, slide gesture signal peak deviation is reduced by integral processing. The random free-sliding gesture signals are parsed by the Transformer algorithm with 96.83 % accuracy. Multi-threaded signal processing improves the response speed and throughput of the system. The accurate real-time gesture recognition for drone flight direction control in extreme environment is realized. This human machine interface can facilitate development of next generation intelligent remote interactive platforms.