Silent Spies

Robotic vacuum cleaners can spy on our conversations. When your robotic vacuum cleaner is busy doing household chores, be aware that it may be picking up conversations along with dust and dirt. Computer scientists from NUS have demonstrated that it is indeed possible to eavesdrop on private conversations using a regular robotic vacuum cleaner and its built-in light detection and ranging (Lidar) sensor.

The new method, called LidarPhone, turns the Lidar sensor, typically used by robotic vacuum cleaners for navigation, into a laser microphone for eavesdropping on personal conversations. The research team, led by Associate Professor Jun Han from NUS Computer Science, and his doctoral student, Mr. Shriram Sankararaman, was able to reconstruct speech data with high accuracy. Mr. Sankararaman shared, "The proliferation of smart devices, including smart speakers and smart surveillance cameras, has increased opportunities for hackers to track our personal data. Our method demonstrates that confidential data can now be collected using something as innocuous as a household robotic vacuum cleaner. Our work highlights the urgent need for practical solutions to prevent such malicious attacks."

The work was presented at the Association for Computing Machinery's Conference on Embedded Networked Sensor Systems (SenSys 2020), where the team received the second-place award.

How the attack works:The core of the LidarPhone attack method is the Lidar sensor, a device that emits an invisible scanning laser and creates a map of its surroundings. By reflecting lasers off common objects, such as a trash bin or a food package located near a computer speaker or a television sound panel, an attacker could obtain information about the original sound, which caused the surfaces of the objects to vibrate. Using signal processing and deep learning algorithms, speech can be reconstructed from the audio data, potentially revealing confidential information.

In their experiments, the researchers used a regular robotic vacuum cleaner with two sound sources. One was a human voice reading out numbers played through the computer speaker, and the other source was musical clips from TV shows played through the television sound panel. The team collected over 19 hours of recorded audio files and passed them through deep learning algorithms that were trained to either match human voices or identify musical sequences. The system was able to detect spoken digits, which could represent a victim's credit card or bank account number. The musical clips from TV shows could potentially reveal the victim's viewing preferences or political orientation. The system achieved a classification accuracy of 91% for speech digit reconstruction and 90% accuracy for classifying musical clips. The researchers also experimented with common household materials to test how well they reflect the Lidar beam and found that the sound recovery accuracy varied for different materials. They discovered that glossy polypropylene bags were the best material for reflecting the Lidar beam, while glossy cardboard performed the worst.

Preventing the attack:To prevent unauthorized use of Lidars, the researchers recommend users not to connect their robotic vacuum cleaners to the internet. The team also suggests that Lidar sensor manufacturers include a mechanism that cannot be disabled to prevent the internal laser from triggering when the Lidar is not rotating.

BUY ROBOTIC VACUUM CLEANER