We describe a method for determining normatively optimal task allocations between human operators and automation in an Optionally Piloted Vehicle (OPV). Our method can be used statically during design or dynamically during flight to advise or restrict task transfers between agents. We use detailed human performance and workload models created via the U.S. Army's IMPRINT tool, but then transform them into state networks from which Markov Decision Processes (MDPs) can be generated and solved for a policy. The policy will dictate how tasks should be allocated between various performance methods in order to optimize long-term expected utility. This approach builds on, but greatly extends, work by Kirlik (1993). We employ a much higher fidelity human performance model, and permit much more flexible human/automation interactions. We provide multiple examples of the use of this approach to address task allocation questions relevant to the design and performance of an OPV.
展开▼
