Safety in Embodied AI: A Survey of Risks, Attacks, and Defenses
A survey and curated catalog of more than 500 papers on safety risks, attacks and defenses in embodied AI, spanning perception, cognition, planning, interaction and agentic systems.
This project accompanies an academic survey that proposes a unified safety framework for embodied AI, covering more than 500 works across perception, cognition, planning, interaction and agentic systems, including adversarial attacks, backdoors, jailbreaks, prompt injection and autonomous-driving safety.
The project page presents the paper corpus as a browsable catalog with sorting, search and community upvoting, alongside a paper reader and a separately screened feed of the most relevant recent work, backed by a self-hosted service that also tracks a visitor counter.
It is oriented at researchers and practitioners studying embodied AI safety rather than at shipping a runnable tool.