Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces an agentic gamification framework for developing robot safety policies through adversarial synthetic scenarios, where a Red Team generates hazardous situations and a Blue Team refines safety measures. This iterative adversarial game effectively uncovers high-risk edge cases that traditional simulation methods may overlook. The approach combines classical risk modeling with modern learning techniques, offering a scalable solution for enhancing safety in physical AI systems operating in complex environments.
Adversarial scenario generation reveals critical safety edge cases that traditional methods miss, paving the way for more robust robot safety policies.
In this work, we propose an agentic gamification framework for hazard-informed learning of robot safety policies through synthetic scenarios. We model scenario generation as an adversarial game between two agents: a Red Team that explores the space of potential failures by constructing hazardous situations, and a Blue Team that incrementally refines safety policies to prevent them. This iterative process enables efficient discovery of high-risk edge cases that are unlikely to be captured through random simulation or manual enumeration. By combining classical risk modeling with adversarial scenario generation and modern learning paradigms, this work provides a scalable pathway for embedding safety into Physical AI systems operating in complex real-world environments. The paper describes ongoing work. The contribution is a problem formulation and a proposed solution architecture.