Event

Real-time recognition of five interaction states achieved with gecko-inspired adhesive interface

1 reports · 1 sources

Researcher Eduart Abdullahu proposed a real-time interaction sensing framework that embeds capacitive sensing in a gecko-inspired dry adhesive interface, capable of recognizing five interaction states (idle, proximity, contact, engaged, disengaged) in on-orbit servicing grasping. Among six lightweight machine learning models, Random Forest achieved the highest offline classification accuracy at 95.65%; when deployed online in combination with a rule-based state machine, accuracy increased from 78.02% to 93.14%, with a macro F1 of 0.9343 and an inference period of 53 ms. The data were collected experimentally and underwent augmentation processing.