Fixed robots have proven themselves in industrial environments, but autonomous and semi-autonomous robots must operate in changing and often rugged climates. This requires more robust sensors and programming techniques. Environmental analysis takes additional computing power.
Luckily, improvements in these areas are leading to greater availability of robots in commercial, industrial, and military settings. Yet robot development remains a mixture of "secret sauce" and heavy-duty consulting.
For years, production-line robots have built and painted cars, sorted and assembled small components, and tested computers. Though not autonomous or mobile, these production robots remain very useful.
Remote-control vehicles often are misrepresented as robots, but they possess many similar characteristics as semi-autonomous robots. These vehicles have proven valuable in search and rescue missions and long-range exploration, such as NASA's Mars Rover.
Though robots offer solutions to a host of problems right now, mass production is limited to very few commercial robots. In many cases, the problems are more difficult than most engineers think, until they learn about the difficulties with sensors, software limitations, and lack of standards. Robotic technology is improving rapidly, but it's still early in the learning curve.
Autonomous robots are difficult to build and program. Nonetheless, that hasn't stopped companies like Sony from creating successful products such as Aibo (Fig. 1). And this little pup, which costs about as much as a laptop computer, doesn't just sit on a table waiting for someone to type.
The latest Aibo has an 802.11b wireless link for remote connectivity. Users can remotely view pictures taken with Aibo's camera. Aibo's voice-recognition system understands over 100 words, and it will perform tricks by showing the Aibo a card courtesy of a new image-processing system. In addition, Aibo's software is more sophisticated. For instance, it can play soccer with its pink ball. It can also locate its own charging station. Motor and gear noise is lower, and it uses a new tactile touch sensor system. Increased processing power, memory, and new sensors make this all possible.
CELL PHONES AND DSPs
Better sensor technology, new batteries, and high-performance, low-power processing systems are significantly impacting the emergence of autonomous robots. Most of these advances don't stem from robotics R&D, but rather from commercial products like camera phones and portable multimedia devices with low-cost DSPs.
Even technologies used in Segway's Human Transporter (HT) are useful in robotics. DSPs and tilt sensors have found their way into many mobile robots as well as the HT. The Segway Robotic Mobility Platform (RMP) is simply a short version of the HT model. It supports a 100-pound payload with a zero turning radius that's handy for robot navigation. Top speed is eight miles per hour with an 8- to 10-mile range.
Some technologies, like the RMP, come prepackaged and ready to go. Others, such as camera modules for cell phones, require additional software. While robot-specific technology is rare, this growing area should begin to produce robot projects. For now, dealing with the mechanicals and interfacing them to the software that will analyze inputs and control outputs remain major hassles.
Now, robots are meeting military needs. Northrop Grumman's Global Hawk Unmanned Air Vehicle (UAV) has delivered reconnaissance information under remote and autonomous control. Flying actually simplifies a robot's job because there are fewer things to avoid, and radar sensors often can detect hostile areas.
Unmanned Ground Combat Vehicles (UGCVs) are proving more difficult in the ongoing research at Carnegie Mellon's National Robotics Engineering Consortium, the Defense Advanced Research Projects Agency (DARPA), and companies like Boeing. DARPA's recent Grand Challenge started with 21 teams who took on a course of over 250 miles. Though none made it over eight miles, next year's competition is expected to crown a winner.
Commercial products experiencing success, such as iRobot's PackBot, take on a more conservative problem (Fig. 2). The robotic platform, replete with articulated treads, comes in standard configurations for exploration. Another version features a manipulator suitable for explosive and ordinance removal. Its use literally spans the globe, from local police departments to Afghanistan.
Products at the other end of the spectrum include iRobot's Roomba (Fig. 3). This little robot shows how judicious choices in the approach to a problem can greatly simplify the solution. Roomba contains only a single 8-bit microcontroller running a behavior-based program. The interface features only three buttons. It can clean most rooms' low-pile carpet or hard floors using a semirandom movement. The trick is not to get stuck, which is where a series of simple sensors comes into play.
Joseph Jones, Roomba's designer, explains that cost, safety, and durability were key design issues. Features were discarded if they were too complex or expensive. For example, a competitor has a robot vacuum system that can detect an object before coming in contact with it. A typical demonstration has the robot stop before it knocks over a wine glass. The Roomba uses bump sensors and would know the glass is knocked over. On the other hand, how many people keep fragile objects on the floor?
He recommends keeping it simple. Don't design a robot because it is "cool." Design it to solve the required problem.
When looking for robot software, understand that there's no such thing as a standard now. Individual platforms like the Sony Aibo and Dr. Robot's HR6 Humanoid Robots have their own software framework that won't be found elsewhere on the planet (Fig. 4).
On the other hand, today's emphasis on diversity has led to many different design approaches. Haipeng Xie of Dr. Robot explained how the HR6 distributes work between the robot and a PC via a wireless link. Processors in the robot handle resident behaviors for movement and software running on the PC handles higher-level functions.
This approach reduces the computing power on the robot, diminishing power consumption. Boosting the performance of the PC allows more complex audio and visual analysis from the robot's sensors while eliminating the storage bottleneck that exists in most standalone robots. Wireless links also make it significantly easier to debug untethered robots.
Evolution Robotics is working toward a standard robotic platform. Its Evolution Robotics Software Platform (ERSP) targets autonomous robot navigation and vision. Jack Weissberg, director of Embedded Systems at the company, explains that ERSP uses standard sensors like USB cameras for input and, with its "secret sauce," delivers localized positional information to an application that controls the rest of the robot. ERSP uses automatic landmark recognition to perform its magic. One platform that Evolution's software runs on is the Evolution Scorpion (Fig. 5).
Evolution's solution represents both the solution and the problem with robotic software today. ERSP provides a robust, comparatively low-cost approach that works on a wide range of platforms. It can handle multiple sensors and cameras with a device-driver-style, XML configuration interface. On the other hand, its behavior-based system is tightly coupled to the rest of the system. It features a multilevel system with a high-level, task-oriented, goal-seeking system needed by most sophisticated robotic applications.
This means that all or a major part of a robotic solution will come from one company or be developed in-house. So it's not surprising that most robot companies consider consulting a major portion of their work.
HARDWARE IS HARD WORK
As Scotty of Star Trek once said, "Damage control is easy. Reading Klingon—that's hard." For robots, reading Klingon is more of a software and sensor issue. Motors and movements are actually easy.
Control of actuators and motors is well understood. DSPs offer improved power consumption and system control and enable the use of less expensive motors for many robot applications. With additional improvements in this area, robot designers won't be changing the approach to problems, only the magnitude.
Currently, articulated robots can jump and run. This is courtesy of robotics research in this field, as well as the improvements made in motor control.
The next step involves work with shape memory alloys (SMAs), which has moved from research into practice. SMA wires are often called muscle wires. These nickel-titanium alloy wires (commercially known as "Nitinol") contract and expand when current is applied and removed. Actuators using SMAs are finding their way into semi-robotic applications like prosthetics, too.
LEARNING ABOUT ROBOTS
Interest in robotics remains high in the commercial as well as educational areas. Yet courses on robotics haven't been as numerous as C++ or SQL classes. Nonetheless, every major engineering university offers courses and ongoing research.
Robotics work isn't restricted to higher academia. Competitions like FIRST (For Inspiration and Recognition of Science and Technology) bring robot research and competition to high schools (Fig. 6). The FIRST Lego Mindstorm competition targets even younger technologists.