A hierarchical control system (HCS) is a configuration wherein a collection of devices and controlling software are organized in a hierarchical tree structure. When the connections within the tree structure are established using a computer network, the resulting hierarchical control system can be classified as a networked control system.

It's common for a human-made system with complex behaviour to be set up in a hierarchy. For instance, the organizational chart of superiors, inferiors, and channels of organizational communication is one of the distinguishing characteristics of a command hierarchy. Similar organizational structures are used in hierarchical control systems to split decision-making authority.

Component of the hierarchy

A linked node in the tree represents each component of the hierarchy. While sensations and command outcomes move up the tree from subordinate to superior nodes, commands, tasks, and goals to be accomplished move down the tree from superior to subordinate nodes. Additionally, nodes can communicate with one another via messages. A hierarchical control system's two defining characteristics involve its layers.

• Each higher tree layer takes longer to plan and execute than its bottom tier.
• Higher layers plan and coordinate lower levels' local duties, goals, and feelings, but they don't usually override them. In a hybrid intelligent system, sub-symbolic reactive layers are the lowest. Higher layers with loosened temporal constraints can reason from an abstract world model and plan. Hierarchical control systems plan well with hierarchical task networks.

In addition to artificial systems, it has been proposed that animal control systems are structured hierarchically. Perceptual control theory posits that an organism's behaviour serves as a mechanism for regulating its perceptions. This theory suggests that the control systems of the organism are structured hierarchically, aligning with the construction of their perceptions.

Application

The hierarchical paradigm emphasizes planning, particularly motion planning, and is one of the robotic paradigms. Since the 1980s, NIST has dedicated resources to studying the potential of computer-aided production engineering. A five-tiered strategy for controlling production was created using the AMRF's expertise in automated manufacturing. DARPA funded studies in the early 1990s to create networked intelligent control systems, with potential uses including military command and control. 

The Real-Time Control System (RCS) and Real-time Control System Software developed by NIST are examples of general hierarchical control systems that have found applications in operating a manufacturing cell, a robot crane, and an automated vehicle.

Conclusion

Subsumption architecture is a popular approach to AI development in behaviour-based robotics. This architecture can break down complex, intelligent behaviour using layered organization and a series of "simple" behaviour modules. More complex concepts are implemented at higher levels, each layer serving a different purpose for the software agent (or system). For example, while deciding whether or not to proceed, the eat-food layer will factor in the decision made by the lowest obstacle-avoidance layer. Instead of being predetermined by an overarching planner, some behaviours may be activated only when sensory inputs indicate they would be suitable.

Furthermore, reinforcement learning has acquired behaviour in a hierarchical control system where each node can learn to enhance its behaviour with experience.

Image source: Unsplash