EEPW论坛

Judith M. Myerson (jmyerson@bellatlantic.net)
Systems Architect and Engineer
July 8, 2003

Systems architect and engineer Judith Myerson explains the ins and outs of wireless robots: their components, their shortcomings, and how they can interact in a competitive or cooperative team within professional environments. Learn how smarter robots can relieve us of the most tedious -- and dangerous -- tasks.

Wireless robotics can make our lives easier, safer, and more enjoyable. While the bulk of wireless robotics is currently in research mode, robotic technologies are becoming more available. In this article, I describe what wireless robotics is and how it is evolving. First, I'll discuss the components that make up a robot. Then, I'll talk about how the robot works alone or with others in a competitive or cooperative team in office, industrial, and military environments. Finally, I'll provide a list of their shortcomings -- but not without suggesting development solutions to overcome them.

The sum of a robot's parts
Robots come in different body parts, either as a single mobile machine or assembled as a humanoid. For example, a wireless robotic arm can gently lift breakable items (eggs) or relieve humans of monotonous, precise tasks, such as sweeping. Robots can also perform dangerous tasks. Sea divers, for instance, use a wireless underwater robotic machine to retrieve hard-to-reach items in loosely embedded rocks, which could potentially fall apart when disturbed (see Darrick Addison's "Introduction to Robotics Technology" in Resources).

With a face or without, all robots come equipped with one or more mechanical devices and sensors (primarily vision and pattern recognition). Some sport artificial intelligence (from low to advanced), and others speak in nearly natural-sounding voices. Many robots have wireless capabilities to allow them to interact with external environments, perhaps guided by human agents.

But all robots have some type of a brain in any body part. A robot's brain is usually comprised of microcontroller systems (MCUs), like the CPUs of your PCs and laptops. According to Addison, MCUs are slower, smaller, and require much less memory than their CPU counterparts. These robotic brains, however, are not for general use; they likely perform specialized tasks such as face or fingerprint recognition.

A robot and its environment
How a robot interacts with its environment depends on how it learns while performing an assigned task. According to David Bruemer and Ron Rujam in "Behavior-Based Robotics: Robot Learning" (see Resources), a robot that must collect rock samples from a distant planet "may have plenty of time to learn new strategies for traversing the planet surface." This means the robot must be able to adapt its behavior to various situations it cannot foresee -- without human intervention.

In contrast, this same robot might not have enough time to learn how to respond in a battlefield, where quick human responses are crucial to winning the war. That robot must get military training on a site away from the battles to learn new motions to perform various tasks.

Not all robots can be fully autonomous; some cannot handle or learn new behaviors in unexpected or unfamiliar situations. These robots must rely on adjustable autonomy (see "Adjustable Autonomy for Marsupial Docking" in Resources), allowing human agents to intervene for assistance and guidance.

Three Cs: cooperation, competition, and control
Wireless robots can interact as a cooperative team (marsupial robots) or as a competitive team (for example, legged soccer player robots) with varying degrees of control.

Marsupial docking
Marsupial robots act as a family. Like a kangaroo, a large mother robot carries one or more smaller child robots. The mother provides a docking platform to load and unload her "kids." Should the mother's eyesight (her computer vision) begin to fail, for example, her kids will guide her back to their home using wireless sensors and adjustable autonomy. Marsupial robots are useful in search and rescue missions, performing repetitious tasks, and handling hazardous materials.

In an industrial scenario, a parent robot acts as the supervisor, while the smaller robots assume the role of office assistants in various departments. When the day begins, the supervisor marches to a designated spot and rolls out the office assistants. The supervisor then commands them when and how to do the repetitive tasks for their human colleagues and when to return to the supervisor -- for an overnight rest or other duties. Marsupial docking is still in research mode, but this scenario is possible using wireless technologies.

Robotic soccer players
The problem with marsupial robots is that they work together as a cooperative team without an element of competitive drive. As shown by the RoboCup games, friendly robotic competition is a good example of "the intelligent multi-agent cooperation and control in a highly dynamic environment." (see RoboCup 2001 Overview in Resources), meaning that competition not only concerns cooperation, but also control.

The RoboCup games take place once a year on a green-carpeted field. What makes the game so interesting is that each player has one type of vision: global or local, with adjustable autonomy capability. Those with global visions carry an overhead camera like a hat and communicate wirelessly with a PC off the field to "identify and track robots as they move around the field" (RoboCup 2001 Overview). Those with local visions already have attached sensors, allowing them to process information onboard or transmit it wirelessly to the PC for further processing.

Making robots smarter
Bruemer and Rujam divide the implementation of robot learning into four groups: artificial neural networks, reinforcement learning, evolutionary learning, and learning by imitation. The first type takes a supervised approach, while the next two allow unsupervised approaches (to some extent). The fourth approach, learning by imitation, uses a biologically-inspired development paradigm to enable the robot to emulate.

The problem with these approaches, however, is that they are all task-oriented. I suggest adding motion-oriented learning: To learn a new motion, a mobile office assistant robot must be knowledge-oriented and able to utilize advanced expert systems on small chips embedded in its brain.

You can also classify robots that can reason and make decisions in an office. You could implement rudimentary reasoning capabilities, for example, by applying fuzzy logic to problems of AI, robotics, and expert systems. Research work in these two classification types is currently underway.

Roam if you want to
If you have more than one robot in an office, the robots might cooperate with each other depending on the way they are built and the degree of sophistication of their pre-programmed learning behaviors. All robotic office assistants are equipped with wireless technologies to relieve roaming workers and business travelers of repetitive tasks.

The following describes how you can use wireless technology to communicate with or even call up a robotic office assistant (the one who can speak, sit, look for files, and so on) from two perspectives:

• Roamers: Using the IBM High Rate Wireless LAN PC card, for example, a worker can roam the office with his ThinkPad while giving instructions to his robotic office assistant (up to 100 meters from the LAN, see http://www.wi-fi.org/). This card works with the IBM High Rate Wireless LAN Gateway, part of Easy Wireless Essentials. You could add up to four wired users, including the office assistant.
• Mobile office users: With Nokia mobile communications and IBM e-business products and services, a business traveler can use any Wi-Fi-certified LAN card for public wireless connectivity at airports, hotels and other public buildings to access the Internet and his corporate intranet wirelessly (see "Using Ada-Based Robotics to Teach Computer Science" in Resources). Wi-Fi certification ensures the interoperability of wireless devices using the IEEE 802.11b standards.

While many robots are driven by AI software, the current trend promotes open-industry robot control software to benefit reuse in various robotics projects. RoboML (Robotic Markup Language) is "designed to support communication between human-robot interface agents, as well as between robot-hosted processes and between interface processes..." (Addison). However, SOAP, which transports RoboML, is not always interoperable.

While some industrial robots can perform an office task the same way every day, we might see an emerging learning-based robot that can adapt its behavior based on some information it receives -- to learn new skills and motions each day. According to Bruemer and Rujam, "for each task, a designer must decide precisely what should be learned, when learning should occur, the computational means to implementing, and how much priority knowledge should be supplied."

Defeat the defects
While you might enjoy the marvels of a mobile robot in your offices, you must consider some of the pitfalls associated with them. However, don't be discouraged; here are some ways you can overcome the pitfalls:

Pitfall 1: Vibration. Many robots vibrate due to their fast motors. Vibration can cause a robot to move from its designated location and not perform its task successfully. Solution: Test the robot for stability and reliability and change parts when necessary.

Pitfall 2: Overload. A robot can overload when it underutilizes tasks with respect to its speed. Solution: Change the design to properly balance out the tasks and the speeds needed to perform those tasks.

Pitfall 3: Development. Using C and C++ to program your robot might cause problems with arrays and pointers. Solution: Use Ada to help you apply software-engineering principles and better track the progress in each life cycle stage, from concept to deployment. (See "Using Ada-Based Robotics to Teach Computer Science" in Resources.)

Pitfall 4: Security. Hackers can exploit a robot's vulnerabilities and turn it into a weapon or completely disable it. Solution: Install safeguards to counter these vulnerabilities.

Pitfall 5: User Expectations. Users sometimes unrealistically expect that robots can make decisions. Solution: Educate users on what robots can do (learning tasks and motions) and cannot do (reasoning and making decisions).

Pitfall 6: Viruses. PCs that are wirelessly communicating with robots might contain viruses. Solution: Install anti-virus programs and present security awareness programs for programmers and users.

Pitfall 7: Dimensions. The mouse can only provide two-dimensional positioning for a three-dimensional robot. Solution: Use a program that lets you create and view a three-dimensional robot or a game of robots in progress.

Pitfall 8: Batteries. Battery power can wear off and slow down, for example, the speed of the arm and the agility of its fingers. It can cause the fingers to accidentally drop an egg, creating a mess on the floor. Solution: Implement power management software that can sound an alert when the robot reaches a low power level; redesign the robot to lengthen the battery life; or switch to better-grade batteries.

Pitfall 9: Analog to Digital Conversion. When using wireless technology for robot communication, human agents and external objects are impacted by the technological limitations in converting analog inputs into digital outputs. For instance, the conversion process can distort soft sounds in voice recognition (analog "hear" converted to digital "held"). Solution: Take advantage of conversions not significantly affected by technological limitations.

Pitfall 10: Bandwidth. Bandwidth problems during wireless transmissions might cause a sensor, say vision, to produce jittery outputs. Solution: Optimize bandwidths so the human eye cannot notice the jittering.

Preparing for robotic living
This article is one of the first to discuss wireless robotics from an integrated approach -- the next step in the evolution to higher levels of robotic technologies. As we prepare for a future with wireless robots, it's important to consider the security risks involved. Wireless networks are guaranteed to become more complex as we drive robotics to maturity. To mitigate the risks and correct security flaws, risk assessment and security monitoring programs must be put in place. Then, the workforce can truly benefit from wireless robotics.

Resources

• "Introduction to Robotics Technology," (developerWorks, September 2001) by Darrick Addison, covers mechanical design, sensory systems, electronic control, and robotic software.
• "Behavior-Based Robotics," by David Bruemer and Ron Rujam, discusses the development of intelligent behavior in mobile robots. (See the "Robotic Learning" section.)
• "Adjustable Autonomy for Marsupial Docking," by Anron Gage, Robin Murphy, and Brian Minten, discusses those robots that are not fully autonomous, but need human agents to assist them.
• The The First RoboCup American Open was held at Carnegie Mellon University April 30-May 4, 2003. Both two-legged and four-legged robots participated.
• Learn more about RoboCup 2001 from The Robotics Institute, Carnegie-Mellon University.
• The Wi-Fi Alliance is a non-profit international association that certifies interoperability of wireless LAN products, based on IEEE 802.11 specs.
• Download the Easy Wireless Essentials from IBM.
• "Using Ada-Based Robotics to Teach Computer Science" discusses motivations for choosing Ada over other high-level languages, and a particular implementation over possible alternatives.
• "Nokia and IBM Join Forces to Expedite Public Wireless LAN Roll Outs" is the press release describing the effort by IBM and Nokia to offer technologies for mobile office use.
• "Wireless Data Virtualization," (developerWorks, May 2003) also by Judith Myerson, serves up definitions of four types of virtualization, traces their evolution, and explains how to build virtualization stacks.