Robots: The Next Generation Embedded Computing System

By Lawrence Ricci

As Microsoft’s Chairman Bill Gates has boldly predicted, computers will change our lives more in the next 10 years than they have in the last 20. One type of these more influential computers we can expect to see is the intelligent and autonomous sensor platform. In a word: Robots. The development of robots requires no breakthrough in underlying silicon or software technology - the technology is already here. The development of ‘smart phone’ generation cell phones, has created the CPUs and sensors needed to drive an incredibly diverse range of robotic products. The key is imaginative use of the hardware and software.

Far from being the stuff of science fiction movies, robots are already part of our lives. The Roomba home vacuum cleaner robot from iRobot, for instance, has sold more than one million units. There are robotic sensors monitoring highway traffic flow and providing security in warehouses. Smart sensor platforms called StaBots are measuring global warming by drifting with the movement of European glaciers. High-end robotic toys such as RoboSapiens and Aibo “play” with us to provide entertainment and education. University programs and robotics clubs are everywhere. Even human-like robots are beginning to appear. Some Japanese companies, concerned about Japan’s declining population, have started to develop fully anthropomorphic robots able to work alongside human beings using the same tools and machinery.

One significant use of robot technology is the creation of Unmanned Aerial Vehicles (UAVs). These systems combine sensor technology, flight navigation and control, and communications capability to provide an ability to remotely investigate and monitor large geographic areas. UAVs are being used to monitor forests for fire, rivers for flood, volcanoes for eruption, crops for pests or disease, and snow fields for avalanche.

A major source of UAV development is in the context of military missions. On the battlefield, the value of robots is already inestimable. Convoy vehicles can and will be robotically driven. Manned convoys- or even individual soldiers- can have protective swarms of UAVs overhead, able to detect threats and even intercept them. Robots could guide aircraft or land vehicles on decoy missions to draw out and disclose defenses or ambush. This is not ‘blue sky’ speculation; DoD funding programs are already in place to make this happen as shown in Figure 1.

Figure 1 � Robot systems such as unmanned autonomous vehicles have been in development for years, and the effort is expected to increase significantly.

Technology Here Today

Such UAV robot designs do not have to wait for ‘miracle chips’ and technology. The technology for developing such applications can be found in the same CPUs and sensors that make up high-end cell phones. Developers just need to glue the available cell phone chips together within an embedded, low power framework. Of course, reconfiguring commercial off-the-shelf (COTS) cell-phone technology, when it comes delivered in chip form, is a bit more difficult than configuring COTS technology using bus cards and “plug-and-play” standards. However, it can be done and done within typical project time and budget limits. Companies like Applied Data Systems can turn cell phone technology into an ECP (Embedded Computing Platform) focused on a particular application, with tremendous cost and performance benefits.

Cell phone technology is significant to robot development for two reasons. One is that it marks a quantum price performance leap past traditional embedded technology for these target applications. As detailed in Table 1, a high performance cell phone costs about $650 before the carrier subsidies that make them so cheap on the merchant market. Even at the true cost, however, the cell phone is far less expensive than an equivalent system build up from traditional computer components. Further, the ECPs made from cell phone components can be very small, ideal for UAV and StaBot applications.

The relatively low cost of cell phone technology makes practical the development and deployment of disposable robots costing less than $2000. This allows disposable UAVs to be deployed in largely autonomous ‘swarms’ of hundreds - even thousands - of units, all communicating over a wireless point-to-point network to coordinate a broad mission in a high-risk environment such as a battlefield.

Even more than cost, however, the advantage of adopting of cell phone technology in the embedded space is the performance improvement it provides over traditional microcontrollers. As the table shows, the processing power available as a function of power consumption in cell phone CPUs offers a 50x advantage over the hardware used in traditional embedded systems. With the addition of other, converging technology developments, even more performance enhancement can be expected.

Reapplying Cell Phone Technology

CPUs such as the Intel® PXA27x processor family have all the features needed to provide this fifty-to-one performance improvement in mobile robots such as UAVs. All that is needed is reapplication of the hardware resources. The embedded 4 MegaPixel image register on the PXA270 processor, for instance, was aimed at the needs of camera phones. Yet it offers the high speed data access needed to make functions such as target recognition possible. Many CCD, CMOS and IR image sensors are available making hyper-spectral imaging solutions possible, simplifying the problems of target recognition. Sensors can be applied in fish-eye, panorama or even binocular applications which would allow close-in maneuvers such as target interception or collision avoidance.

Similarly, the on-board DSP/MAC, designed for telephone automatic speech recognition, can be used to monitor vibration and acoustic signatures in the environment. The interfaces to civilian radio wide area networks (WANs) can instead connect to JTRS (Joint Tactical Radio Service), Iridium, or another tactical network. The typical Bluetooth™ personal area network interface can be replaced by low power sensor nets like Zigbee, (http://www.zigbee.org/) or a specifically designed UAV point-to-point network such as MeshNet (http://www.meshnetworks.com/).

Other converging technologies can further enhance the utility of these robots. UAVs may, for instance, fly within the broadband IP ‘cloud’ of a Metro Area Network (MAN) such as the 802.11.16 WiMax™, using the network to establish point-to-point communications among the UAVs. Location and orientation technology from GPS chips can augment time-of flight signal data from other robots in a point-to-point environment to allow coordinated actions within a UAV swarm. Low cost MEMS (micro electromechanical systems) originally developed for image stabilization of camcorders can provide six-access inertial orientation data for the robots.

In low power robot applications, such as the StaBots, the low end of the power envelope can achieve a micro-amp sleep region by including a microcontroller on the board under integrated power management. For systems with complex image or signal analysis needs, the addition of an FPGA can extend the available compute power to TerraOps or further.

None of this technology is breakthrough; it only requires integration of available components. Yet the cumulative result can be astounding. With 20 sensors on each UAV just microseconds away from software response routines, and a thousand or more UAVs each with a GigaOp or more of embedded processing, the potential performance of future robotics systems extends well into the supercomputer realm.

Software Support Also Available

Applying so much sensor and processor power to a mission may seem a daunting software development task, but it can be accomplished within object-oriented software frameworks already available. Real-time CORBA (common object request broker architecture) implementations exist, simplifying the sharing of data and resources. Also, Windows CE.NET offers both real-time response and managed code in a shrink-wrapped package with source code and ubiquitous support.

These object-oriented methods decompose both vertically and horizontally the problem of developing software for large UAV swarms. For instance, developers can compare a dozen kinds of image analysis/target recognition software from companies such as Leading Edge Design (http://www.leds.com) within a stable XML/SOAP framework. At the same time, other development team members can perfect various options for maneuver (takeoff, land, patrol, avoid, etc) independently of sensors analysis.

Concerns over network security and remote provisioning of these cell phone-like robotic systems are also readily handled. Cell phone security technology is years ahead of the desktop environment because cell phone carriers have been so cautious with their networks. Companies such as Odyssey software (http://www.odysseysoftware.com/) and Senforce (http://www.senforce.com/) currently offer proven packaged solutions for these important tasks.

The software problem will be rendered soluble not just because of existing technology, but because of the emergence and empowerment of 3rd party software suppliers. The wide availability of a low cost sensor/CPU platform for robotic applications will allow many companies to attack the problem. Perhaps even more important, full deployment of this technology will create new classes of robots and StaBots. Emergence of broadly deployed robotic platforms would then encourage the development of a 3rd party software industry, accelerating the development of AI (artificial intelligence) functions and increasing the utility of ‘swarm’ configurations.

Clearly, robots are coming. Cell phone technologies are providing ECPs with tremendous cost and performance enhancements. At the same time, there is tremendous need for robotic systems within both defense and civilian applications that will continue growing. Meanwhile, underlying sensor and image processing technology continues to advance and commercial successes in all the necessary technologies have already emerged. The convergence of all these factors is making robots the next generation of embedded systems.

Table 1 � The cost and feature set of cell phone electronics represent a dramatic improvement over the more conventional standard board-based embedded system.

---

Lawrence Ricci is the Business Development Manager for Applied Data Systems, an Associate member of the Intel® Communications Alliance (ICA) and a leading provider of low-power RISC-based embedded systems. Mr. Ricci is a Microsoft Embedded MVP and a participant in ITAFF (Information Assurance Technical Framework Forum) and SWWG (Secure Wireless Working Group).