Tactile sensing artificial finger skin: equivalent multi-round dense sampling training strategy for network...

Area of Science:

• Robotics and Artificial Intelligence
• Materials Science and Engineering
• Optical Sensing Technologies

Background:

• Fiber Bragg grating (FBG) arrays offer robust, compact tactile sensing ideal for robotic fingers.
• Current data-driven models require extensive calibration, increasing time and workload.
• Developing efficient calibration methods is crucial for practical FBG tactile sensing.

Purpose of the Study:

• To present an equivalent multi-round dense sampling training strategy (EMRDS-TS) for FBG-based tactile sensing.
• To reduce the calibration time and experimental workload associated with FBG tactile sensors.
• To enable accurate simultaneous position and force estimation using FBG arrays.

Main Methods:

• Developed an EMRDS-TS combining a deterministic trend model with a conditional generative adversarial network.
• The trend model captures the FBG array's wavelength response to position and force.
• A coordinate attention-based haptic perception demodulation network (CA-HPDNet) was trained on reconstructed data.

Main Results:

• EMRDS-TS achieved performance comparable to 8-12 rounds of dense sampling.
• Position prediction (0-30 mm) yielded a Mean Absolute Error (MAE) of 0.4228 mm and R² of 0.9790.
• Force demodulation (0-9.81 N) resulted in an MAE of 0.2265 N and R² of 0.9749.
• The strategy's effectiveness was validated on a second, identically configured sample.

Conclusions:

• The EMRDS-TS significantly reduces calibration requirements for FBG tactile sensing.
• The proposed method enables accurate and efficient haptic perception for robotic systems.
• This strategy offers a practical solution for deploying FBG-based artificial skin.