Towards Reliable Autonomous Agents

We are interested in producing reliable autonomous robots that can operate for extended periods of time in uncertain, dynamic environments. We have been developing methodologies and software tools to facilitate this, including the Task Control Architecture and probabilistic methods for representing and reasoning about uncertainty. The aim is to incrementally produce reliable behavior by adding (perhaps automatically) execution monitors and exception handlers that can detect when expectations are being violated and react appropriately. We have been developing and testing these ideas on a number of mobile robot platforms, including a six-legged planetary rover, a wheeled Lunar rover, and an office-navigation robot. Introduction Reliability is a key aspect of autonomous agents, especially those that operate in uncertain and dynamic environments. They need to know their capabilities and limitations, when they are getting into trouble, and how to recover from exceptional situations. We ha..