Slip detection for compliant robotic hands using inertial signals and deep learning

Area of Science:

• Robotics
• Sensor Fusion
• Machine Learning

Background:

• Object slip during grasping is a common challenge in robotics.
• Passive compliance in robotic hands can lead to unpredictable slip events.
• Sensing fingertip motion and vibration is crucial for detecting slip.

Purpose of the Study:

• To investigate the use of inertial measurement units (IMUs) for detecting slip events in passively compliant robotic hands.
• To determine if orientation changes and slip-induced vibrations can serve as reliable slip indicators.
• To develop and validate a machine learning model for automated slip detection.

Main Methods:

• A tendon-driven, underactuated robotic hand was used to perform 195 manipulation trials.
• Data from an inertial measurement unit (IMU) at the fingertip was collected.
• Motion-tracking data was used for automatic labeling of slip and non-slip events.
• A convolutional neural network (CNN) was trained to detect slip events from IMU data.

Main Results:

• The trained CNN successfully detected slip events from IMU data.
• Slip detection remained effective even with other disturbances present.
• The model demonstrated robustness by performing well on data from a different gripper and object.

Conclusions:

• IMU data, capturing orientation changes and vibrations, is a viable source for slip detection in robotic grasping.
• CNNs can be effectively trained to identify slip events from sensor data.
• The developed slip detection method shows promise for real-world robotic manipulation tasks.