Robotics is a lot harder than it looks. But that's what makes it so challenging. Rolling, flying, and walking robots are hard enough to build. Now try creating a robot in human form. Yet the state of the art continues to improve by leaps and bounds, including a few robots that are starting to do just that.
The problems associated with humanoid robots are varied and many. That's why so many projects tackle only a subset of these functions without attempting anything as complex as the android Lieutenant Commander Data on Star Trek: The Next Generation.
In fact, challenges such as voice, obstacle and image recognition, emotional response, and eye-hand coordination are daunting tasks all by themselves. So when ST:TNG's Lieutenant Natasha Yar asks, "You are fully functional, aren't you?" the answer would be "Not yet."
Complex tasks like running or balancing on one foot are now common, though. These movements aren't completely lifelike, but they are smoother. High-performance processors, hardware such as digital cameras to provide situational awareness, and improved artificial intelligence (AI) are enabling researchers to create very lifelike robots. Just like high-end gaming systems, the results may be almost indistinguishable from the real thing within a constrained context.
The complexity of these robots is growing significantly with human robotic research. Systems typically have tens of degrees of freedom (DoF) as fingers and toes come into play for balance, interaction, and picking up the tab. This is especially true of robots that run or have facial expressions.
Robots Running Around Honda's ASIMO (Advanced Step in Innovative Mobility) gets by with just 26 DoF (and 26 servos), so heading out for a jog is a typical affair (Fig. 1). A few of these ASIMO research vehicles cost millions.
Still, ASIMO has been a flexible development and research platform. Its hip joints allow it to pivot even while it's walking or running. This permits the robot to shift its center of gravity, which is very handy for biped movement.
Walking and running essentially are nothing more than controlled falling. Honda's research has yielded a smoother gait and more natural movement, as ASIMO uses algorithms that can look ahead and anticipate any necessary corrections. Honda calls this real time technology iWalk (Intelligent Walking). ASIMO can handle inclines over 30° as well as stairs of different sizes.
As a semiautonomous robot, ASIMO can take general directions (such as "run ahead 10 steps") much like NASA's Mars Exploration Rover, which receives general instructions about where to move to and then moves to the desired location on its own. ASIMO's master uses a wireless PC.
ASIMO is constructed of lightweight materials, including a magnesium alloy. Its backpack houses a 40-V nickel-metal-hydride battery and the main computer. It can operate for about half an hour. It's only 120 cm high and weighs 52 kg. And, it can run over 3 km/hour.
ASIMO's hands have an opposable thumb and four fingers that can grab small objects that weigh up to 1 lb. It can even shake hands. A pair of cameras and audio input and output are located behind its faceplate. It can handle image and voice recognition, though its level of autonomous response is limited. What Honda has accomplished is impressive, but ASIMO's simple faceplate is rather expressionless. It also provides opportunities for others to provide this type of support.
Making Robotic Faces Attempts to get a robot to express itself have been going on for a while. Cynthia Breazeal and a host of graduate students in the Humanoid Robotics Group at the Massachusetts Institute of Technology Artificial Intelligence Lab have completed some groundbreaking work, starting in 1993 with Kismet (Fig. 2).
This anthropomorphic robot was designed to show emotions when it interacts with people. The Kismet project concentrated on the face. Its high-level perception, motivation, behavior, motor skill, and face motor systems ran on on four Motorola 68332 microprocessors running an application developed in L, a multithreaded Lisp environment. This setup was tied to a network of nine QNX-based PCs to handle speech recognition and analyze vocal intent.
Kismet had a 15-DoF face whose expressions could mirror emotional states such as happiness, sadness, or anger. Each ear had two DoF. And, the robot could manipulate its eyes, eyebrows, and lips.
Reprints
)
