Sensor spoofing attacks

Dear everyone,

I have come across two sensor attacks that have been recently discussed in academia. Unlike most others, these specific attack vectors demand my serious attention. I am particularly interested to know if the ArduPilot software is susceptible to these sensor attacks. If it is, I would like to understand how we can prevent such attacks at the control software level as well as hardware measures. Details are below.

  1. Intentional sound noise causes resonant on IMUs
    Motivating Example - YouTube
    Details: UnRocker

  2. Electromagnetic interference (EMI) disturbs sensor readings on SPI/I2C buses
    EM injection experiments on a hovering drone - YouTube
    Details: Paralyzing Drones

These two attacks potentially make the drone crash on the ground. Thus, I would greatly appreciate your insights and opinions on these sensor attacks. I am curious to know if we should consider them a significant concern or not. Thank you!

Best regards,

1 Like
  1. This attack should be countered by using dissimilar IMU units with different resonance frequencies. Since EKF innovation drastically increases if only one IMU is affected EKF switchover should happen. I wouldn’t expect this to be significant threat unless directed acoustic system like LRAD is used.

  2. This may be quite problematic for non hardened boards and also may be significantly harder to test for. Proper shielding of flight control system and use of EMI resistant communication for communication with peripherals should help mitigate this kind of an attack.
    I would like to see cube tested like that as they have mostly metalic casings.

1 Like

Yeah I’d be interested to see these tests carried out with Arducopter with well-configured filters and multiple IMUs - as you say.

EMI shielding is well understood and usually only employed for military or commercial aviation use (to different extents) - this is really just an additional cost not associated with hobby grade equipment.

1 Like

I stumbled upon this thread unexpectedly and became intrigued. I’d like to add an additional type of sensor attack to this discussion.

It appears that optical flow sensors have been successfully deceived by light projection, although these attacks were only effective indoors due to the limited intensity of light.

Should we remain apprehensive about this attack if we assume they can project light at a high intensity?


Details

Kind of doubt there will be a discussion. Maybe if it was posted on Discord. People will talk about anything there. You could start a new channel.

1 Like

@dkemxr Thanks for the answer. Unfortunately, it seems that I cannot create a channel on Discord. I just asked creating the new channel on Discord.

This vector of atack seems extremely unlikely to be viable in outdoor environment.

1 Like