Reconciling Design Flexibility and Efficiency

The last section demonstrated that the lowest cost approach to the implemention of desired behaviors in many-robot systems will depend sensitively on both the relative costs and capabilities of candidate sensors and effectors and the detailed requirements of the intended application. The system designer therefore needs a highly flexible integration paradigm that can quickly and easily support the "evolution" of robots to exploit the differing combinations of subsystems needed to address the requirements of an expanding diversity of applications, as well as to accommodate continuing rapid advances in component technologies.

As Flynn argued in [FLYN87], the real payoff from tiny robots will come from "tight" physical integration of components: size, weight and power requirements will have to be minimized in order to minimize cost. Cables, connectors, and duplicated resources will be primary targets for elimination. The ultimate goal is monolithic fabrication of complete robots in a fashion analogous to (and perhaps nearly identical to) the manufacture of integrated circuits. Electronic Design Automation (EDA) tools for the development of ASICs offer a useful model of what can be done to support the development of complex, highly integrated systems, and the process of extending these tools to MEMS processes and designs has begun.

Even realizing the dream of monolithic MEMS fabrication of micro-robots, however, will not solve all our integration problems. Many applications will continue to require macro-sized robots, and the growing diversity of relevant sensor and effector technologies implies an ever growing requirement to integrate more or less "off the shelf" sensors into robots, for singly-deployed robots as well as for many-robot systems. The problem is to reconcile the need for lightweight, compact, inexpensive, high bandwidth, "tight" integration of subsystems with the need for flexibility in configuring systems from diverse subsystem elements.

Efficient Integration Architecture

The word "architecture" is often used in the robotics community in the sense of "intelligent control architecture"; i.e., as a mechanism for implementing desired behaviors of effectors based on inputs from sensors. Examples are Brooks's subsumption architecture [BROO86] and Albus's RCS [ALB88]. An integration architecture, on the other hand, is a mechanism for gluing predefined subsystems together into a coherently functioning whole. (Yet another use of the word is in the sense of implementation architecture, as a style of designing a system and building it -- one central processor vice a number of smaller ones.)

Unfortunately, sensor subsystems on the market exhibit a dizzying variety of interfaces for communicating data, control, and status: analog, discrete binary digital, PCM digital, parallel digital, serial digital, servo control, relay closure, and mixtures of the above. Sometimes a vendor decides to provide a "complete subsystem solution" by adding a microcontroller and providing the user with an asynchronous RS-232 link. Products available with an RS-232 interface include video cameras, pan/tilt mechanisms, laser rangefinders, and radar track processors. RS-232 has several serious deficiencies, however: (a) it is a point-to-point link which connects one device to one other device, so that using many devices requires many serial ports; (b) it is slow, imposing long message latency time and providing limited bandwidth; (c) it is expensive to use since it is necessary to parse the incoming character stream to extract meaningful data; and (d) it is asynchronous, so that the arrival of a complete message can only be determined by a succssful message parse, and it is difficult to reliably detect a link failure.

Fortunately, LAN technology is now readily available to solve many of the basic communications deficiencies of RS-232, connecting multiple nodes with high speed synchronous communications. The OSI Reference Model [ISO84] supplements basic LAN technology with a protocol framework which could be employed to represent robot-specific software constructs. Off-the-shelf sensor or effector subsystems with LAN interfaces are still, however, rare. Some integrators of large robotic (and telerobotic) systems have provided a LAN interface for each subsystem by incorporating an additional processing element (variously termed a Front End Processor, Network Front End, Interface Processor, or Intelligent Communications Interface [GAGE85b]). Clearly, this approach is at variance with "optimal" per-system cost-minimization; however, continuing advances in VLSI technology allows the implementation of microcontroller elements costing only a few dollars in volume production. A number of candidate network architectures for distributed control have emerged in recent years to address such application areas as building control (including the "Smart House" concept) and integrated automotive electronics [RAJI94], and the IEEE and NIST have undertaken a joint effort to develop a communications standard for "smart sensors" [BRYZ94]. Echelon's LonWorks architecture in particular seems to offer a number of valuable features to support robotic integration: typed network variables, a robust protocol stack, a choice of diverse LAN media, and protocol extensibility.

Up to Many Robot Systems

Robotics home page

Please address all questions and comments to: robo-web@spawar.navy.mil
Last update: 1 December 1998.