June 3, 2024 at 01:56AM
A team of researchers demonstrated how they turned a Baidu autonomous vehicle (AV) into a weapon using a low-cost method involving metal foil and colored patches, which could hide a target vehicle from the AV’s perception system. This attack, published in a conference last week, exploits vulnerabilities in multi-sensor fusion systems and could be executed with drones. Baidu’s robo-taxi operations have expanded in China.
Based on the meeting notes, the key takeaways are:
– A team of researchers from various universities was able to turn an autonomous vehicle (AV) into a deadly weapon by tricking its multi-sensor fusion system using a “low cost” and “easily fabricated” method that involved strategically arranging metal foil and colored patches on cardboard to hide a target vehicle from the perception system of a victim AV.
– The attack leveraged mmWave reflection to affect radar perception and color patch misrepresentation to affect camera perception, thus compromising all three sensing modalities.
– The researchers suggest that the attack could be carried out with drones or by mounting the adversarial object on the front vehicle and disguising it as an advertisement, allowing for high stealthiness and flexibility.
– While the attack used Baidu Apollo platforms, it is theorized that the attack strategy could be applied to other multi-sensor fusion systems.
– Baidu’s robo-taxi operations have expanded across China, charging for autonomous rides in its Apollo Go cabs and now operating in more than ten Chinese cities, with an expected profitability for its robo-taxi wing next year.
Let me know if there’s anything else you need assistance with.