SAFE: A Novel Approach for Real-Time Detection of Slip and Fracture in Robotics

Manipulating delicate objects poses significant challenges for robots, particularly in understanding the state of the objects they handle, such as detecting slip or fracture. The authors present SAFE, a sensing approach utilizing two passive polyvinylidene fluoride (PVDF) acoustic sensors combined with motor proprioception, which operates independently of visual input or prior knowledge of material properties.

The sensors are integrated into a compliant Fin Ray gripper, and a HistGradientBoosting classifier processes a 79-dimensional feature vector to determine the object's state (normal, slip, or fracture). In experiments, SAFE achieved an Alert-F1 score of 0.884 with minimal confusion between slip and fracture states.

Additionally, an adaptive grasp controller utilizing this detection method operates at 104 Hz on a Jetson Orin Nano, achieving a 91.3% success rate across 46 trials with various object types. Notably, it also succeeded 82.4% of the time with novel objects not seen during training, showcasing its effectiveness in failure-aware grasp control without the need for object-specific calibration.