Weight-Quantized SqueezeNet for Resource-Constrained Robot Vacuums for Indoor Obstacle Classification

With the rapid development of artificial intelligence (AI) theory, particularly deep learning neural networks, robot vacuums equipped with AI power can automatically clean indoor floors by using intelligent programming and vacuuming services. To date, several models have been proposed to distinguish objects between cleanable litter noncleanable hazardous obstacles. Unfortunately, these existing focus entirely on accuracy enhancement object classification, little effort has made minimize memory size implementation cost models. As a result, require far more space than typical vacuum provide. address this shortcoming, paper aims study find an efficient model that achieve good balance classification usage (i.e., cost). In work, we propose weight-quantized SqueezeNet for vacuums. This classify litters from obstacles based image or video captures Furthermore, collect videos pictures captured built-in cameras use them construct diverse dataset. The dataset contains 20,000 images ground-view perspective dining rooms, kitchens living rooms various houses under different lighting conditions. Experimental results show comparable around 93% while reducing at least 22.5 times. More importantly, footprint required our is only 0.8 MB, indicating run smoothly resource-constrained vacuums, where low-end processors microcontrollers are dedicated running algorithms.