Protecting Smart Homes from Unintended Application Actions

Many smart home frameworks use applications to automate devices in a smart home. When these applications interact in the same environment, they may cause unintended actions which can lead to a safety violation (e.g., the door is unlocked when the user is not at home). While recent efforts have attempted to address this problem, they do not capture complex app behaviors such as: 1) timed behavior and user inputs (e.g., a door can remain unlocked for a long time because of a lock-door app that locks the door after x duration, if x is set too large.) and 2) interactions between devices and the environment they implicitly affect (e.g., water sprinklers cannot be turned on if the water supply is off). Hence, prior work leads to many false positives and false negatives. In this paper, we present PSA, a practical framework to identify safety intent violations in a smart home. PSA uses parameterized timed automata (PTA) as an expressive abstraction to model smart apps. To parse these apps into PTA, we define mappings from smart app APIs to equivalent PTA primitives. We also provide toolkits to model devices, environments, and their interactions. We evaluate PSA on 86 apps in the Samsung SmartThings IoT ecosystem. We compare PSA against two state-of-the-art baselines and find: (a) 19 new intent violations and (b) 35% fewer false positives than baselines.