Abstract:
Multiple agents operating in a shared environment can interfere with each other’s ability to reach their goals. One of the approaches to address this issue is enacting a social law — a set of rules that restricts some possible behaviors of the agents. A social law that ensures that each agent can achieve its goal, regardless of what the other agents do, is called robust.

In this work, we strive to uncover deeper theoretical foundations of social laws in the context of STRIPS-like environments. We start by pointing out and correcting a mistake in the previous work. Furthermore, we establish the semantics behind the waiting mechanism and connect it to the agent’s sensing capabilities. Using these insights, we also derive methods for faster robustness validation. Finally, we reason about the social law robustness in a reactive environment, getting one step closer to real-world scenarios.