Control architectures

A control architecture is a software framework that manages both sensor and actuator systems, thus enabling the AUV to perform a user-specified mission. There are basically three different approaches to control architectures: deliberate, reactive and hybrid systems (Arkin, 1998)(Valavanis et al, 1998)(Ridao et al., 2000)(Russell & Norvig, 2003). Deliberate (hierarchical) architectures are based on planning using a world model. They allow reasoning and making predictions about the environment. Data flows from sensors to the world model, which is used together with a predefined set of mission goals to plan new actions to be undertaken by the actuators (Fig 10). This "Sense-Plan-Act" scheme is suited for structured and predictable environments, and its well-defined, tightly-coupled structure simplifies the process of system design and verification. However, problems arise when trying to maintain an accurate, real-time world model in the complex, dynamic and only partly observable environments typical for underwater operation. System response may then be too slow or erratic.

Fig. 10. The deliberate control architecture is based on world modelling and planning

Fig. 10. The deliberate control architecture is based on world modelling and planning

Reactive (also called behavioural) control architectures are based on the simple "Sense-React" scheme and typically involve neither world model nor planning. Basically, a group of parallel behaviours act independently generating outputs to the actuators according to their input sensor data (Fig. 11). A mission is normally described as a sequence of phases each with a set of active behaviours. A "Transit" phase could for instance have the active behaviours "Avoid collision", "Go to waypoint" and "Error handling". The system's global behaviour emerges from the coordination of the elemental active behaviours and its interplay with the environment.

Fig. 11. The reactive control architecture is based on independent, parallel behaviours

Since each behaviour pursues its own goal, behaviours may issue contradicting actions, making behaviour coordination an important part of the reactive system. The different methods of behaviour coordination can be divided into a cooperative and a competitive approach (Carreras et al., 2000). In the cooperative approach, the behaviours vote on desired actions (e.g. the desired direction to move). The resulting action can then be either the one with the most votes, or a weighted average of the behaviour outputs. The votes can be multiplied by a gain value according to the priority of the voting behaviour. It is aso possible to implement negative votes, or even veto votes, to avoid making the vehicle move into forbidden or dangerous regions.

The competitive approach to behaviour coordination has a single behaviour in control of the vehicle at any given time. An example is the well-known subsumption architecture (Brooks, 1986), where higher-prioritized behaviours will override (subsume) lower-priority behaviours when activated. For instance, an "Avoid collision" behaviour will subsume a "Go to waypoint" behaviour when the vehicle is approaching an obstacle. Competitive behavioural coordination provides more robustness than cooperative coordination since safety behaviours may act undisturbed in dangerous situations. As only one behaviour acts at a time, the competitive approach is also more modular and easier to tune. Cooperative behaviour coordination, on the other hand, tends to give more optimized paths. This can be seen by considering that the "Avoid collision" behaviour will be able to manipulate the vehicle path at an earlier time, making the resulting path smoother and less shortsighted. The cooperative approach also allows for working on several goals simultaneously, which is not possible with the competitive approach.

Reactive control is sensor-driven and appropriate for making low-level decisions in realtime. However, reactive control rarely yields a plausible solution at the global level, because global control decisions depend on information that cannot be sensed at the time of decision making. For such problems, deliberate control is more appropriate. Furthermore, it can be hard to predict the overall vehicle behaviour when reactive control is used. Consequently, most AUV architectures use reactive techniques at the lower levels of control with deliberate techniques at the higher levels. Such architectures are called hybrid architectures. The most popular is the three-layer architecture, which consists of a reactive layer, an executive layer and a deliberate layer (Gat, 1998)(Brutzman et al., 1998) as seen in Fig. 12.

Fig. 12. The hybrid control architecture combines reactive and deliberate control.

The executive layer serves as a supervisor, accepting directions from the deliberate layer and sequencing them for the reactive layer. Similarly, it receives status and processed sensor data from the reactive layer and sends invocations to the deliberate layer. Decision cycle is usually on the order of milliseconds, one second and minutes for the reactive, executive and deliberate layer, respectively (Russell & Norvig, 2003).

Was this article helpful?

0 0
Learn Photoshop Now

Learn Photoshop Now

This first volume will guide you through the basics of Photoshop. Well start at the beginning and slowly be working our way through to the more advanced stuff but dont worry its all aimed at the total newbie.

Get My Free Ebook

Post a comment