Modeling Hydra Behavior Using Methods Founded in Behavior-Based Robotics

6.3. Constituent behaviors 25Release type, where a higher-priority behavior enables the output of a lower-prioritybehavior to pass through the switch.6.2.1 Applicability clauseThe AC defines, for each behavior, whether the output from the behavioral unit shouldbe active or not. An AC can be situation-driven, i.e. related to a goal state, where thepresent state (resulting from the agent’s action) causes the AC to be either true or false. Asan example, consider a robot that turns only when an obstacle appears in front of the robot(and only until the object is no longer in front of it). The AC can also be event-driven. Anevent is characterized by a very brief (point-like) occurrence, whereas situations typicallyare extended intervals of time [14]. An event-driven AC is of a set/reset type, where acertain event triggers the AC, and some other criteria resets it.6.2.2 Transfer functionThe TF determines the actual action (output) of the behavioral module, given the currentmotivational state. The output from the TF is only passed out from the behavioral modulewhen the AC is true.6.3 Constituent behaviorsUsing the CSA, modeling each of the constituent behaviors consists of two steps, namely(1) defining an AC, and (2) defining a TF. The architectures of a behavioral module isillustrated in Fig. 6.3.As described in Section 4.1, there are several different methods by which behaviors canbe generated. In the case of Hydra, the constituent behaviors are relatively simple, andthus implementation by means of hand-coding was used in most cases. As a complementto hand-coding, biologically inspired methods such as RNNs and EAs were also used.A brief description of these methods is given in Appendices A and B, whereas specificimplementation details for the resulting behavioral model of the simulated Hydra arestated in Chapter 7.