Search papers, labs, and topics across Lattice.
1
0
3
2
Adversarial prompts can hijack VLM-controlled robots with a success rate of up to 29%, exposing a critical vulnerability in their operational integrity.