Embedded Files
DISCLAIMER
DISCLAIMER
All experimental results from lidar output in the paper and appendices are the "raw" point-cloud output. Practically, (semi-)autonomous vehicles or vehicles equipped with ADAS further processes point-cloud to perform noise filtering, point clustering, classification, etc. Therefore, even if the same raw lidar outputs are provided, their system-wide impact can drastically vary according to automakers' implementations.
UPDATE: Many errata of the contents have been corrected (9th Nov 2019).
Error Correction w.r.t. PRT of the PLD
Error Correction w.r.t. PRT of the PLD
With regard to the smallest supported PRT of the PLD (OSRAM SPL PL90), the paper states that it cannot be reduced less than 100 us. However, this seems not to be true because Cao et al. (Adversarial Sensor Attack on LiDAR-based Perception in Autonomous Driving, ACM CCS 2019) succeeded to inject pulses with a much shorter PRT than this (2.304 us). The reason we came up with 100 us was that its datasheet stated its maximum pulse width and duty cycle are 100 ns and 0.1% respectively, from which 100 ns * 1/0.1% = 100 us could be derived. However, the minimum PRT cannot be derived from the maximum pulse width and duty cycle but possibly from the minimum pulse width and the maximum duty cycle.
Appendix A: Video Captures of Experiments
Appendix A: Video Captures of Experiments
Light Exposure to Weak Light Source - Playback of raw packet capture: Blinding by Weak Light Source
Light Exposure to Strong Light Source - 1st Scene is the playback of raw packet capture: Blinding by Strong Light Source
Lidar Obliquely Exposed to Strong Light Source - Playback of the raw packet capture: Blinding by Obliquely Incident Light Source
Lidar Spoofing of Multiple Moving Fake Dots - Playback of raw packet capture: Delay Change (5Hz, last)
Lidar Spoofing While Changing Spoofer PRF - Playback of raw packet capture: Pulse Frequency Change (5Hz, last)
Lidar Spoofing of Fake Dots Closer than the Spoofer - 1st Scene is the playback of raw packet capture: Spoofing Closer than Spoofer (5Hz, dual, 9.0418kHz, 30 cycles)
N.B. For the ease of viewing induced dots, only laser5 was displayed.
Lidar Outdoor Spoofing - 1st scene is the playback of raw packet capture: Outdoor (5Hz, dual, laser1, 1.9m)
Appendix B: Raw Packet Captures of Experiments (in .pcap)
Appendix B: Raw Packet Captures of Experiments (in .pcap)
N.B. The list below redirects to external download links. Files can be viewed with Veloview by opening .pcap files and selecting the lidar type as VLP-16.
1. Raw packet captures of Blinding attacks
2. Raw packet captures of Spoofing attacks
[Subject of Exp] ([Update Rate], [Lidar Mode], [Pulse Repetition Freq.], [# Cycles], [Lidar~Spoofer Distance]).
Otherwise noted, laser 5 (firing order in VeloView) was spoofed, the distance between the lidar and the spoofer was 5.3m, and # cycles was 5 & 10 for 5 & 10Hz of update rate respectively.
Experimental setup for blinding by obliquely incident light source
Appendix C: Curved Glass Refraction/Reflection Experiment
Appendix C: Curved Glass Refraction/Reflection Experiment
To check whether oblique incidence on curved glass can really make the light come from directions other than that of the light source, we prepared a transparent acrylic cylinder on the right side. Its outer diameter & thickness is 105mm & 3mm respectively. N.B. This is almost the same size as VLP-16.
Then, we put the cylinder on VLP-16 as if it were an additional lidar on top of it.
We obliquely aimed the cylinder and turned on the laser module as shown in the figures below. N.B. The red figure is taken by the IR viewer, a modified digital camera.
IR viewer directly facing light source inside the cylinder. The figure on the right side was taken by the IR viewer
As in the figures below, we could capture the light comes from the direction not that of the of the light source. The figure on the right side was captured from the IR viewer.
We confirmed that this is not from outside the glass, but from the glass itself. If the light were from the outside, this should be seen when the IR viewer is outside the cylinder. However, as in the figures below, it is not. This light can only be seen when the IR viewer is inside the cylinder, which means it is from the glass itself. Below the figures are from a video captured while moving the IR viewer in & outside the cylinder.
Appendix D: Receiver Comparator Circuit Diagram
Appendix D: Receiver Comparator Circuit Diagram
Appendix E: Complementary Materials for Equipment Used
Appendix E: Complementary Materials for Equipment Used
N.B. The list below redirects to external links.
30mW, 905nm laser module
800mW, 905nm laser module
Focusable 0.8w 905nm 800mw IR Infrared Laser module Diode Module
Pulse Laser Diode Driver
Google Sites
Report abuse