Flexible sensor system for recognizing objects during robotic grasping

A new piezoelectric tactile sensing system can identify everyday objects in real time with high accuracy.

Researchers developed a flexible, distributed piezoelectric tactile sensing system. This system helps with early-stage object recognition during robotic grasping and manipulation.

The study tested the system using six everyday objects. They used three recognition approaches: feature-based classification using support vector machines (SVMs), raw-signal classification using 1D convolutional neural networks (1D-CNNs) and convolutional recurrent neural networks (C-RNNs), and spatiotemporal heatmap classification using a 2D-CNN.

Tactile sensing provides information about object properties. A spatiotemporal heatmap is a method used for classification. The system was tested with different sensor counts and temporal window sizes.

The best method used spatiotemporal heatmaps and a 2D-CNN. It achieved F1-scores of 0.97960.9796, 0.98460.9846, and 0.99030.9903 using 16, 24, and 32 sensors. Increasing the number of sensing elements improved recognition performance.

The optimal configuration used 32 sensors and a 100-ms temporal window. This setup was deployed on an embedded edge platform. It achieved an end-to-end recognition latency of approximately 101 ms and energy consumption of 63.8 microjoules per inference.

Key facts

  • The system achieved F1-scores up to 0.99030.9903 with 32 sensors.
  • Recognition latency was approximately 101 ms from contact onset.
  • Energy consumption was 63.8 microjoules per inference.

All sections

Edition