Mobile Robot

Summary and Comparison

We have seen four architectures, of which two (layered architecture and blackboard) are very specific and give precise indications as to the components expected in a robot. The other two (control loop and implicit invocation) define no functional components and concentrate on the mechanisms.

Specificity is helpful for getting a grasp on the basic abstractions and tasks involved in an autonomous robot. It would be interesting to research the value of a TCA architecture (which is the most powerful in its mechanisms) combined with a functional decomposition of robot tasks (planning, sensor integration, ...).

Other hybrid architectures have been proposed. The NASA/NBS Standard Reference Model for Telerobots (NASREM) [Lumia90] can be seen as a combination of the control loop and the layered architectures.

The NASREM Architecture

The layers from top to bottom are defined by the time frame in which they perform their tasks. Seen from this perspective, the architecture is a hierarchy of control loops with increasingly tighter response time constraints.

The layers from left to right represent the functional abstractions.

To conclude, the table summarizes the strengths and weaknesses of the reviewed software architectures.

	Control Loop	Layers	Impl. Invoc.	Black Board
Task Coordination	+-	-	++	+
Dealing with uncertainty	-	+-	+-	+
Fault tolerance	+-	+-	++	+
Safety	+-	+-	++	+
Performance	+-	+-	++	+
Flexibility	+-	-	+	+

Strengths and Weaknesses of Robot Architectures

Updated Halloween 95 by Mary Shaw
Comments to maintainer