Here - Agents Lab - University of Nottingham

utilized to provide solutions for different tasks according to their real-time requirements.For example some BDI-based control components can be devotedto implement simple event handling tasks to guarantee real-time reaction andresponse to critical events and other BDI-based control components can be usedto implement complex goal-based deliberative behaviors with relaxed real-timeproperties. In addition to facilitate the development of real-time control systems,distributed control architectures are also effective in dealing with software developmentcomplexity [7] and allow for easier and more efficient use of paralleland distributed computing resources whenever available.To facilitate the development of real-time BDI-based distributed control systems,suitable methodologies and tools are needed to develop real-time BDIbasedcontrol components and analyze and guarantee their real-time properties.Also a dedicated architecture is required to provide necessary mechanisms forreal-time communication between and coordination of distributed BDI-basedcontrol components. State-of-the-art research on design of robotic control architecturespresents interesting ideas for coordination and synchronization ofdistributed control components, which can guide the development of real-timedistributed BDI architecture. The following briefly introduces two representativesof such approaches.4.1 T-REXT-REX [30] control architecture comprises a set of coordinated concurrent controlloops named reactors, and a functional layer encapsulating a robot low-levelfunctionalities. Reactors maintain their own view of the world and have theirown control functionalities and temporal properties (i.e. lookahead window fordeliberation and deliberation latency), therefore allow for partitioning a controlproblem in both functional and temporal horizons. T-REX has a centraland explicit notion of time which allows execution of all reactors to be synchronizedby an internal clock, ensuring the current state of the control system tobe kept consistent and complete. The unit of time in T-REX is a tick, definedin external units on a per application basis. A deliberation time for each reactorin T-REX is bounded by its own deliberation latency, which is defined as anumber of ticks. When a deliberation requires more than one tick, it should bedefined as a proper sequence of steps to allow for interleaving synchronization(i.e. information exchange) in each tick.4.2 ContrACTAnother example of distributed control architectures is ContrACT. The programmingmodel of ContrACT [26] decomposes a robot control software into aset of controllable modules. Modules are independent real-time software tasks,which, depending on their types, use different communication models such asblocking/non-blocking and publish-subscribe/request-reply to communicate witheach other. Some modules are reactive to events they receive, named asynchronous,and others are executed periodically, named synchronous . There is46