Abstract
In this paper, we develop a dust detection sensor system capable of identifying types of dust for an improvement of a robot vacuum cleaner. The dust detection sensor system is composed of a set of infra-red sensors: a single transmitter and multiple receivers. Given the fixed amount of light transmitted from the transmitter, the amount of light coming in multiple receiver sensors varies, depending on the type and density of dust that is passing between the transmitter and the receivers. Therefore, the type of dust can be identified by means of observing the change of the amount of light from the receiver sensors. For experiments, we use two types of dust, rice and sesame, and validate the effectiveness of the proposed method.
1. Introduction
Most commercial robot vacuums currently operate with a binary understanding of debris: they detect either the presence or absence of dust. While some high-end models utilize acoustic or simple optical sensors to gauge debris volume, they lack the granularity to distinguish between different materials, such as fine dust, heavy grains, or pet hair. This limitation prevents the robot from dynamically optimizing its cleaning parameters—such as suction power, brush rotation speed, or travel velocity—based on the specific nature of the mess.
This research proposes a low-cost, reliable infrared (IR) sensor system capable of "fingerprinting" household debris in real-time. By utilizing a multi-receiver array, the system can capture both light transmission and reflection, creating a unique optical signature for different materials.
2. Proposed Sensor System
2.1 Hardware Configuration
The core of the system is a specialized sensor ring designed to be integrated into the intake of a vacuum. Unlike standard single-beam sensors, this system employs one IR transmitter (LED) and five receivers.
As shown in the hardware configuration, the receivers are strategically placed to capture different optical phenomena. Sensors 1 and 2 are positioned to measure light transmission (the degree to which dust blocks the beam), while sensors 3, 4, and 5 are positioned to capture reflected light (how light bounces off the debris). This dual-mode sensing allows the system to distinguish between materials that might have similar transmission properties but different reflective characteristics.
2.2 Data Processing and Classification
The identification process follows a structured pipeline to convert raw optical signals into a material classification.
Because raw sensor data is inherently noisy, the system evaluates two filtering methods: standard Average (AVR) and Moving Average (M/V). The Moving Average filter is particularly effective at smoothing the "spiky" data caused by individual grains without losing the underlying signal characteristics.
For classification, the system uses a Pixel-based Image Similarity method. The time-series data from the five sensors is treated as a 2D grid—essentially an "image" of the debris signature. By calculating the Euclidean distance between a real-time reading and pre-recorded reference templates, the system identifies the material match.
3. Experimental Evaluation
The system was tested using a custom gravity-fed testbed with two distinct materials: rice (large, reflective, obstructive) and sesame seeds (smaller, darker, less reflective). Eight distinct datasets were collected for each material to account for variations in flow and density.
3.1 Results
- Rice Identification: The system achieved a high accuracy of 87.5% across Raw, AVR, and M/V filtered data. The large physical profile of rice makes it easily detectable across all sensor modes.
- Sesame Identification: This material proved more difficult due to its smaller size and lower signal-to-noise ratio. Raw data yielded only 42% accuracy. However, applying the Moving Average (M/V) filter significantly improved performance, boosting identification accuracy to 64%.
4. Conclusion and Future Work
The multi-angle IR receiver array successfully demonstrates that different types of debris can be identified by their unique optical signatures. The use of image similarity techniques on sensor streams provides a robust framework for classification using relatively simple hardware.
Future work will focus on expanding the library of detectable materials to include common household items like hair and lint. Additionally, the researchers aim to miniaturize the sensor array for integration into standard robot vacuum intake throats and adapt the algorithms to handle the high-velocity airflow found in real-world cleaning environments.
Summary updated