Smart Tactile Sensing Systems Based on Embedded CNN Implementations

Abstract

Embedding machine learning methods into the data decoding units may enable the extraction of complex information making the tactile sensing systems intelligent. This paper presents and compares the implementations of a convolutional neural network model for tactile data decoding on various hardware platforms. Experimental results show comparable classification accuracy of 90.88% for Model 3, overcoming similar state-of-the-art solutions in terms of time inference. The proposed implementation achieves a time inference of 1.2 ms while consuming around 900 μ J. Such an embedded implementation of intelligent tactile data decoding algorithms enables tactile sensing systems in different application domains such as robotics and prosthetic devices.

Keywords: convolutional neural network; embedding intelligence; tactile sensing systems.

Figure 1

Block diagram of the tactile sensing system.

Figure 2

Examples of visual (top) vs. pressure (middle) vs. tactile images (bottom) of common objects.

Figure 3

Architecture of the tested model. BaN, Batch Normalization.

Figure 4

Visual representation of the training, test, and validation split using cross-validation.

Figure 5

Example of an image resized for the sticky tape object; the red canvas is shown for illustration, which signifies the original image size (28 × 50).

Figure 6

Learning accuracy for the 3 configurations of the TactNet4model: (a) training; (b) validation.

Figure 7

Comparison of the performance, number of trainable parameters, and FLOPS in the convolutional layers.

Figure 8

Implementation flow.