Lot’s of people have ‘em. Robot sweepers, that chug around cleaning the floor and annoying pets.
When you get one of these, it sets about to “learn” your home, using sensors including Lidar. Which is cool, if you don’t mind a robot exploring your personal living spaces.
This winter researchers at Maryland report a possible hack that could let your robotvac use its Lidar as a microphone that listens to you [2].
Basically, this is the same idea as using a laser to listen to the vibrations of a window or other object. In this case, the laser is the robot’s sensor, repurposed. (It is also necessary to hack a channel to report the data out to the adversary.)
I imagine that in addition to covertly listening, the adversary may be able to steer the robot to follow or select targets.
The authors do not offer much in the way of countermeasures, though vendors could reduce the capabilities of the lidar in ways that reduce the side channel. I’d note that making the robot hard to access remotely would limit the ability to leak information from any source.
I guess the best defense is to not leave the robovac turned on in any sensitive location. And obviously, a bit of black tape over the lidar of the supposedly idle robot would be effective, if invonvenient.
This attack is covert not because you can’t detect the robot, but because you may be accustomed to its presence and consider it harmless. This could be a dangerous assumption if an adversary can covertly access the robot.
This attack also makes me think about even more dangerous attacks based on physically modifying the robot. An adversary could load many kinds of high capability sensors in the bot, far beyond the Lidar attack. It could even load improvised weaponry, to be remotely or automatically triggered. Eeek.
- Kimbra Cutlip, Could your vacuum be listening to you? , in Eureka Alert, November 17, 2020. https://www.eurekalert.org/releaseguidelines/
- Sriram Sami, Yimin Dai, Sean Rui Xiang Tan, Nirupam Roy, and Jun Han, Spying with your robot vacuum cleaner: eavesdropping via lidar sensors, in Proceedings of the 18th Conference on Embedded Networked Sensor Systems. 2020, Association for Computing Machinery: Virtual Event, Japan. p. 354–367. https://doi.org/10.1145/3384419.3430781
Robot Wednesday
Robert McGrath's Blog 