Action selection

Action selection is an architectural approach developed by Pattie Maes [43] that uses a dynamic mechanism for behavior (or action) selection. This dynamic mechanism solves the problem of the predefined priorities used in the subsumption architecture. Each behavior has an associated activation level, which can be affected by the current situation of the robot (gathered from the sensors), its goals, and the influence of other behaviors. Each behavior also has some preconditions that have to be met in order to be active. From all the active behaviors, the one with the highest activation level is chosen for actual execution.

This coordination mechanism resembles very much our bidding approach. In our architecture, each system (or agent within the Navigation system) bids according to the urgency for having the action executed, which is equivalent to the activation level. However, our bidding agents have no preconditions to be met in order to become active, and they are always ready to bid. Another important difference is that behaviors in action selection can influence the activation level of other behaviors, whereas in our approach the agents are totally independent, since an agent cannot influence the bids of another agent.