Search papers, labs, and topics across Lattice.
1
0
3
VLMs in self-driving cars are shockingly vulnerable: a subtle combination of graffiti and foreign-language commands can hijack their behavior without degrading performance on normal tasks.
