Machine Learning Based Characterization of a Waterproofed Soft Multi-Directional Force Sensor

A cost-effective soft multi-directional force sensor can enable next-generation robots to recognize object features by touch, similar to humans. This paper presents a novel approach for force region classification on a soft hemispherical sensor made up of four flex sensors placed $90^{\circ}$ apart. By considering a quarter of the hemisphere due to its symmetry, we effectively divided the sensor into three distinct regions where force can be applied. We used the K-means clustering algorithm to classify the region of the applied force, which successfully categorized all 60 test trials into the correct force regions. The effective range for detection was about 1.5 - 5 N of force. Our experimental results also showed that the sensor responded well to being kept submerged in salt water for 7 days, indicating its suitability for operation in underwater environments. This study opens up possibilities for 3D multi-directional force recognition using the soft sensor, which has potential usage in diverse fields, such as robotics and healthcare, even in challenging underwater environments.