Context
Many autonomous systems operate in closed conditions, where all the variables are known. However, AI systems operating in unconstrained environments, or open systems require new capabilities that allow them to learn and respond when they encounter previously unknown conditions.
Using a smart home as a demonstration model, a methodology was established that allows an AI system to update the reasoning it uses to make decisions, incorporating previously unknown or novel situations.
While existing smart home operating systems are reactive and operate within defined conditions, our model uses a proactive AI operating system that has the capacity to respond to changing conditions. It incorporates intention or goal recognition, which allows it to predict the behaviour of residents (or agents) in the home environment and determine the best way it can help to achieve those goals.
The ability to update its ‘beliefs’ is critical in learning to recognise and respond to goals of family members, pets, or even intruders.
Impact
The smart homes model was developed with funding from an Australian Research Council Grant and successfully showcases emerging capabilities in computational engineering for dynamic systems. The relevant algorithms that provide the foundation for the model can be applied across diverse fields of operation that require AI systems to operate in uncontrolled environments.