Autonomous agents and multi-agent systems have been proved to be useful in several safety-critical applications. However, in current agent architectures (particularly BDI architectures) the deliberation process does not include any form of risk analysis. In this paper, we propose guidelines to implement Tro-pos Goal-Risk reasoning. Our proposal aims at introducing risk reasoning in the deliberation process of a BDI agent so that the overall set of possible plans is evaluated with respect to risk. When the level of risk results too high, agents can consider and introduce additional plans, called treatments, that produce an overall reduction of the risk. Side effects of treatments are also considered as part of the model. To make the discussion more concrete, we illustrate the proposal with a case study on the Unmanned Aerial Vehicle agent.
展开▼
