Japanese researchers build amoeba-inspired robot that navigates terrain blind
A four-legged robot classifies ground surfaces without cameras, using proprioception and reservoir computing.
A University of Tokyo team led by Hyoto Yamaguchi and colleagues has developed a four-legged walking robot that navigates terrain without cameras, using artificial proprioception instead. The system combines a three-axis accelerometer and eight foot pressure sensors with reservoir computing (RC) to classify ground surfaces as flat or rough with high accuracy, even amid dynamic motion.
The robot demonstrates real-time gait adaptation by switching walking styles based on classified ground conditions. The team analyzed sensor contributions, showing how proprioceptive data enables terrain awareness comparable to visual systems but without the computational overhead. The research, submitted to IEEE SCIS ISIS 2026, highlights a scalable approach for blind navigation in robots.
- Four-legged robot uses proprioceptive sensors (accelerometer + 8 pressure sensors) instead of cameras to classify terrain.
- Reservoir computing enables high-accuracy ground classification (~90%+) during dynamic motion.
- Demonstrates real-time gait switching based on terrain conditions without visual input.
Why It Matters
Paves the way for low-power, visually impaired robots in search/rescue or industrial inspection.