Goal-Driven Autonomy in Planning and Acting

To operate autonomously in complex environments, agents must perform actions, sense the environment, and respond to new situations. Traditional approaches face difficulties with incomplete environment models, goal specification, and engineering domain specific control knowledge. We believe goal reasoning will address these challenges enabling agents to better respond to unexpected circumstances and reduce the knowledge engineering burden. We present a conceptual model of goal-driven autonomy (GDA) integrating four reasoning tasks: discrepancy detection, explanation, goal formulation, and goal management. Additional work is required to understand the implications of goal reasoning, and GDA in particular, for improving the robustness of autonomous agents.