Question 1

Why do prompt injection attacks matter for LLM-integrated mobile robots?

Accepted Answer

Prompt injection attacks can exploit vulnerabilities in multimodal prompting, causing the robot to produce incorrect or dangerous navigation decisions that affect both the robot and its surroundings.

Question 2

What system and attack scope does the study investigate?

Accepted Answer

The study deploys an LLM-controlled mobile robot in simulation and investigates its robustness against prompt injection attacks, including how GPT-4o handles such attacks.

Question 3

How do the proposed defense mechanisms improve performance?

Accepted Answer

Implementing robust defense mechanisms via secure prompting strategies improves both attack detection and overall system performance by approximately 30.8%.