5/6/2023 0 Comments Freeciv mapsWhen prediction failures implicate the actions that the agent would use to replan for its current goal, safety goals are made primary instead. General COGENT Architecture this as 'reflective mode' for the simulation. When the cognitive controller is turned on, we refer to Figure 1. The cognitive controller acts in response to differences between the predicted incremental effects of actions and the observed ones. ![]() The simulated cars move forward an incremental amount based on their speed and direction at each tick, and the environmental controller makes observations and generated reactions at each step based on its observations and plans. Figure 2 shows our realization of the general COGENT (Cognitive Agent) architecture for our simplified driver agent operating in simulation. Recognizing expectation failures (surprises) and using reflective reasoning (about potential flaws in its own behavior, knowledge, model of how the world works) to recover and learn from these surprises, should help them improve by being less surprised over time, even though the events themselves are rare. such systems are to be able to gradually gain reliability in more realistic urban driving situations, they must learn from experience but also have the ability to reason about the expectations from prior knowledge and advice they were given to explain how new situations violated those expectations.
0 Comments
Leave a Reply. |