Humanoid Robotics

1. INTRODUCTION

Like never before, technology can bring imagination to life. The question is what will we conceive? For decades, popular culture has been enthralled with the possibility of robots that act and look like humans. We are promised by film, fiction and television that humanoids will cook for us, clean for us, become our best friends, teach our children, and even fall in love with us. So where are they? The forerunners are here already. Recently, the media has covered a surprising number of new humanoid robots emerging on the commercial market. Like many new technologies, these early generations of commercially available humanoids are costly curiosities, useful for entertainment, but little else. Yet, in time, they will accomplish a wide variety of tasks in homes, battlefields, nuclear plants, government installations, factory floors, and even space stations.

Humanoids will exhibit emotion, forge relationships, make decisions, and develop as they learn through interaction with the environment. Robots that can incrementally acquire new knowledge from autonomous interactions with the environment will accomplish tasks by means their designers did not explicitly implement, and will adapt to unanticipated circumstances of unstructured environments. Already, humanoid robots can autonomously perform task decomposition necessary to carry out high-level, complex commands given through gesture and speech. Humanoids can adapt and orchestrate existing capabilities as well as create new behaviors using a variety of machine learning techniques. In fact, some researchers claim to have implemented a first stab at the "seed" which will allow robot intelligence to develop indefinitely. Humanoids may prove to be the ideal robot design to interact with people. After all, humans tend to naturally interact with other human-like entities; the interface is hardwired in our brains. Their bodies will allow them to seamlessly blend into environments already designed for humans. Historically, we humans have adapted to the highly constrained modality of monitor and keyboard. In the future, technology will adapt to us. Undoubtedly, humanoids will change the way we interact with machines and will impact how we interact with and understand each other.

pg. 1

Humanoid Robotics

2. HISTORICAL PERSPECTIVES

Humanity has long been fascinated by the possibility of automata (from the Greek "automatos," acting of itself). In the second century B.C., Hero of Alexander constructed statues, doors and small mechanical animals that could be animated by water, air and steam pressure. By the eighteenth century, elaborate mechanical dolls were able to write short phrases, play musical instruments, and perform other simple, life-like acts. 1 Today, robots are no longer mere curiosities, but have become an indispensable pillar of global industry. We have millions of factory automation robots carrying out complex tasks around the clock. From clockwork, gearfilled devices, we have arrived at lethal instruments of war such as the unmanned military vehicles vividly demonstrated to the world during the 1991 liberation of Kuwait.

From the very beginning, our fascination extended beyond machine automation to the possibility of creating an entity with our own form and function. In Homer's Argosy, the bronze sentinel, Talos, was created and animated by Daedulus to guard the island of Thera. Written sometime around the 3rd century A.D., the pre-Cabbalistic book of Jewish mysticism named the Sefer Yezirah (The Book of Creation) describes how numbers and letters can be arranged to correlate with the four elements of creation (Spirit of God, ether, water and fire) and provide a template for life itself. According to Jewish legend, certain great rabbis used their programming prowess to instill life in an effigy or golem, creating a human-like automaton that could carry out its master's command. Even in myth, humans recognized the uniqueness of their intelligence and the staggering difficulty of replicating it. The legend acknowledged that although the golem could perform simple tasks as it was ordered, it would never possess ru'ah - the breath of life bestowed on Adam in the primordial creation. This myth provides an interesting context for examining the past, present and future of Humanoid Robotics and raises some hard questions. Is human intelligence more than any encoding can capture, no matter how elegant or complex? How should we represent and impart knowledge. What is the best we can hope for?

pg. 2

Humanoid Robotics

3. EARLY ENDEAVOURS

With the rise of the computer, people immediately began to envision the potential for encoding human intelligence into textual programs, but soon discovered that static programs and rulebased logic cannot capture the true essence of human intelligence. Early attempts to create artificial intelligence produced information-processing machines that operated on high-level human concepts, but had difficulty relating those concepts to actions and perceptions in the external world. Estranged from perception and action, such intelligence derived meaning only as an extension of the human creator or user.

In 1973, the construction of a human-like robot was started at the Waseda University in Tokyo under the direction of the late Ichiro Kato. He and his group developed WABOT-1, the first fullscale anthropomorphic robot in the world. It consisted of a limb-control system, a vision system and a conversation system. WABOT-1 was able to communicate with a person in Japanese and to measure distances and directions to the objects using external receptors, artificial ears and eyes, and an artificial mouth. The WABOT-1 walked with its lower limbs and was able to grip and transport objects with touch-sensitive hands. At the time, it was estimated that the WABOT1 had the mental faculty of a one-and-half-year-old child. In 1985, Kato and his research group at Waseda University built WASUBOT, a humanoid musician (WAseda SUmitomo roBOT), developed with Sumitomo Electric Industry Ltd. WASUBOT could read a musical score and play a repertoire of 16 tunes on a keyboard instrument. Since these early successes, the Japanese electronics and automotive industries have played a key role in the emergence of humanoids by creating robots of humanoids by developing robots capable of walking over uneven terrain, kicking a soccer ball, climbing stairs and performing dexterous tasks such as using a screwdriver and juggling. At the present time, we have full-scale humanoid robots that roughly emulate the physical dynamics and mechanical dexterity of the human body.

pg. 3

Humanoid Robotics

4. WHAT IS A HUMANOID ROBOT?

Humanoid Robotics includes a rich diversity of projects where perception, processing and action are embodied in a recognizably anthropomorphic form in order to emulate some subset of the physical, cognitive and social dimensions of the human body and experience. Humanoid Robotics is not an attempt to recreate humans. The goal is to create a new kind of tool, fundamentally different from any we have yet seen because it is designed to work with humans as well as for them. Humanoids will interact socially with people in typical, everyday environments. We already have robots to do tedious, repetitive labor for specialized environments and tasks. Instead, humanoids will be designed to act safely alongside humans, extending our capabilities in a wide variety of tasks and environments.

At present, Humanoid Robotics is not a well-defined field, but rather an underlying impulse driving collaborative efforts that crosscut many disciplines. Mechanical, electrical and computer engineers, roboticists, computer scientists, artificial intelligence researchers, psychologists, physicists, biologists, cognitive scientists, neurobiologists, philosophers, linguists and artists all contribute and lay claim to the diverse humanoid projects around the world. Inevitably, some projects choose to emphasize the form and mechanical function of the humanoid body. Others may focus on the software to animate these bodies. There are projects that use humanoid robots to model the cognitive or physical aspects of humans. Other projects are more concerned with developing useful applications for commercial use in service or entertainment industries.

Defining a humanoid robot is a lot like defining what it means to be human. A humanoid robot is most likely defined as any robot with two arms, two legs and a human-like head. Unfortunately, such a definition says nothing about the ability of this robot to receive information, process it and respond.

pg. 4

Humanoid Robotics

Rather than distinguish humanoids by their physical construction, we choose to identify several complementary research areas that, thus far, have stood out as distinct emphases. Eventually, a fully-fledged humanoid robot will incorporate work from each of the areas below.

Perception:
This area includes computer vision as well as a great variety of other sensing modalities including taste, smell, sonar, IR, haptic feedback, tactile sensors, and range of motion sensors. It also includes implementation of unconscious physiological mechanisms such as the vestibuloocular reflex, which allows humans to track visual areas of interest while moving.

Human-robot interaction:
This area includes the study of human factors related to the tasking and control of humanoid robots. How will we communicate efficiently, accurately, and conveniently with humanoids? Much work in this area is focused on coding or training mechanisms that allow robots to pick up visual cues such as gestures and facial expressions that guide interaction.

Learning and adaptive behavior:

For robots to be useful in everyday environments, they must be able to adapt existing capabilities to cope with environmental changes. Eventually, humanoids will learn new tasks on the fly by sequencing existing behaviors. A spectrum of machine learning techniques will be used including supervised methods where a human trainer interacts with the humanoid, and unsupervised learning where a built-in critic is used to direct autonomous learning.

Legged locomotion:
For humanoids to exploit the way in which we have structured our environment, they will need to have legs. They must be able to walk up stairs and steep inclines and over rough, uneven terrain. The problem is that walking is not simply a forwards-backwards mechanical movement of the legs, but a full-body balancing act that must occur faster than real-time. The best approaches look closely at the dynamics of the human body for insight.

pg. 5

Humanoid Robotics

Arm control and dexterous manipulation:

Around the world, researchers are working on dexterous tasks including catching balls, juggling, chopping vegetables, performing telesurgery, and pouring coffee.

pg. 6

Humanoid Robotics

5. RESEARCH ISSUES

Will humanoid research propel robotics on to great heights, channeling ideas from diverse fields toward an ultimate goal? Or will the quest to model ourselves prove to be a stumbling block, or worse? We may be our best models of intelligence; but then again, we may also be our worst. Although cognitive neuroscience will continue to contribute much to our self-understanding, we by no means fully appreciate the myriad internal processes that actually produce our intelligence.

Problems with past Thinking Robots

In their zeal to make robots "think like humans," early humanoid researchers focused on highlevel cognition and provided no mechanism for building control from the bottom up. Although intended to model humans, most of the systems did not, like humans, acquire their knowledge through interaction with the real world. When situated in the real world, these robots possessed little mastery over it. Even in the fortunate event that sensors could accurately connect internal 'archetypes' to real-world objects, robots could only extend the knowledge thrust upon them in rudimentary, systematic ways. Such robots carried out preconceived actions with no ability to react to unforeseen features of the environment or task.

Building Intelligence from the Bottom-Up

Today, the question for Humanoid Robotics is how best to impart these primitive behaviors to robots. Many researchers find it ineffective to directly hard-code such low-level behavior with imperative languages like C or C++ and instead use a more biologically motivated technique such as artificial neural networks. Artificial neural networks allow a 'supervised' learning approach where a designer trains a system's response to stimulation by adjusting weights between nodes of a network. As it turns out, ANNs fail to capture the recursive power of the human brain. Unlike an ANN where the structure of the network is usually fixed, the brain's highly integrated, well-ordered structure emerges through competition between separately evolving collectives of neurons. Critics argue that ANNs' lack of such an architecture prohibits meta-level learning -- the ability to not only generalize, but also extend acquired knowledge beyond the frontiers of experience.

pg. 7

Humanoid Robotics

Other learning techniques such as reinforcement learning and genetic algorithms have also played a role in modeling various levels of learning. Reinforcement learning can be used as an 'unsupervised,' learning-with-a-critic approach where mappings from percepts to actions are learned inductively through trial and error.

Learning by Doing: Automated Development through Real-World Interaction

The mechanical sophistication of a full-fledged humanoid body poses a devastating challenge to even the most robust learning technique. The more complex a humanoid body, the harder it is to place constraints necessary for productive learning. If too few constraints are employed, learning becomes intractable. Too many constraints on the other hand, and we may curtail the ability of learning to scale. Ultimately, the conventional learning techniques described above are limited by the fact that they are tools wielded by human designers, rather than self-directed capabilities of the robot.

pg. 8

Humanoid Robotics

6. CURRENT RESEARCH PROJECTS

Humanoid Research has already begun to accelerate. While only a few institutions are fully dedicated to the creation of humanoid robots, a host of projects around the world are meeting with encouraging success in particular areas. This section highlights endeavors in legged locomotion, arm control and dexterous manipulation, robot-human interaction, service robots, learning and adaptive behavior, perception, and anthropopathic (emotive) robots.

Legged Locomotion
In 1980, Waseda University developed WL-9DR, the world's first robot to exhibit quasi-dynamic walking. Five years later, Waseda University's Humanoid Research Laboratory teamed with Hitachi Ltd. to develop the WHL-11 (Waseda Hitachi Leg-11) biped, walking robot. Legged locomotion is much easier to accomplish (and much safer to develop and test) on smaller humanoids. The SDR-4X was recently developed by Sony as a domestic robot capable of handling uneven surfaces and stairs on the fly. While other approaches have demonstrated hardcoded walking behavior, the Sony project attempted to create a truly useful robot that can sense depth and distance of objects sufficiently well to be able to walk over obstacles and adjust its gaits on the fly to cope with changing surface heights.

Arm Control and Dexterous Manipulation

The University of Tokyo Department of Mechano-Informatics has developed a humanoid robot, Saika, with a two-DOF neck, dual five-DOF upper arms, a torso and a head. Saika is able to dribble a bouncing ball and catch a thrown ball. It can grope for and grasp unknown objects. Whereas many humanoids are heavy and require large, unwieldy off-board apparatus for actuation, Saika is designed to be lightweight and has almost all the motors built into the arms and torso. The head, torso and arms together weigh less than 17 pounds. NASA has engineered a dexterous humanoid robot that will deploy, maintain and operate a wide variety of shuttle and space-station components. Currently, astronauts must perform a variety of extremely dangerous and costly dexterous tasks. The objective of the Robonaut project is to develop a space robot with dexterous capability exceeding that of a suited astronaut. This will
pg. 9

Humanoid Robotics

reduce response time, high costs and some of the dangers associated with sending an astronaut into space. To accomplish these aims, Robonaut must be able to efficiently assist astronauts by sharing their space and tools. Robonaut will need to respond to natural communication and learn through what NASA JSC calls virtual-reality coaching where the human will effectively take control of the robot and guide it through certain movements and behaviors.

Robot-Human Interaction
For robots to be profitably integrated into the everyday lives of humans within military, commercial, educational or domestic contexts, robots must be able to interact with humans in more meaningful, natural ways. As artificial agents inundate our lives, it will be increasingly important to enable multi-modal, intuitive modes of communication that include speech, gesture, movement, affect, tactile stimulation and context. Body dictates behavior, and if we want a robot to relate with and learn from humans, it must be able to map its body to our own. An ambitious project at MIT is based on the premise that humanlike intelligence requires humanoid interactions with the world. These researchers are developing a robot they call Cog as a set of sensors and actuators that tries to approximate the sensory and motor dynamics of a human body. Cog is equipped with a sophisticated visual system capable of saccades, smooth pursuit, vergence, and coordinating head and eyes through modeling of the human vestibuloocular reflex. Cog responds not only to visual stimulation, but also to sounds and to the ways people move Cog's body parts. By exploiting its ability to interact with humans, Cog can learn a diverse array of behaviors including everything from playing with a slinky to using a hammer.

Service & Entertainment Robots

Human-robot interaction plays a crucial role in the burgeoning market for intelligent personal, service and entertainment robots. Applications include everything from robots that assist the elderly and severely disabled to entertainment robots at amusement parks. Increasingly, robots that can serve as mobile, autonomous tour guides and information kiosks will grace public places. One encouraging example is Minerva, a popular tour guide at the Smithsonian National Museum of American History, which uses a rich repertoire of interactive capabilities to attract people and guide them through the museum. Minerva's facial features and humanoid form have had a profound effect on the way in which people respond to it.

pg. 10

Humanoid Robotics

Learning and Adaptive Behavior

Japanese scientists are searching for learning techniques that can scale indefinitely. At the University of Tokyo, researchers are using a learning methodology they call interactive teaching to give robots the ability to drive their own development. The robot uses Bayesian Networks to map sensor evidence to behavior, and then assigns each mapping a confidence rating. In the beginning stages, confidence ratings are low and the robot must frequently ask a human trainer for help deciding between competing actions. With practice, the robot requires less intervention from the human trainer until eventually a task can be completed autonomously. As the task changes, the robot can again ask for help.

Perception
Humans interact with continuously flowing, diverse stimulation. Likewise, humanoids must have multi-modal perceptual systems that can seamlessly integrate sensors. One way to do this is to allow sensors to continually compete for dominance. At the ElectroTechnical Laboratory in Japan, G.Cheng and Y. Kuniyoshi have developed a humanoid with 24 degrees of freedom, joint receptors with encoders and temperature sensing. The humanoid uses 6 PCs for control of hearing, vision, motor output and integration. The robot itself is lightweight and flexible, allowing it to interact comfortably and safely with humans. Throughout a visual and auditory tracking task, the robot tracks a person by sight and/or sound while mimicking the upper body motion of a person. The focus of the work was in showing that the robot can track people using a multiple sensory approach that is not task-specific and does not need to switch between sensor modalities. The key is that perceptual subsystems necessary for mimicry, tracking, vision and auditory processing should not be thought of as separate tasks and pursued separately, but as essential capabilities that must together contribute to high-utility humanlike behavior.

Anthropopathic Robots
Whereas anthropomorphic robots have bodies that look and physically act like the human body, anthropopathic robots are able to emote. The robots discussed in this section not only perceive and respond to human emotion, but are themselves possessed of an intrinsic emotional system that permeates their control architecture. For these humanoids, emotional state is not merely an outward expression, but can be used to influence the actions and behavior of the robot.
pg. 11

Humanoid Robotics

7. WHAT DOES THE FUTURE HOLD?

While these projects are important steps in the right direction, functional results come slowly. Like the human infants they model, contemporary humanoids are inefficient at most tasks and require intensive training. One of the implications of this research is that to create humanlike adaptability and versatility, it may be necessary to introduce an element of human frailty and inconsistency. We have already manufactured machines with the ability to perform a task in exactly the same way, time after time. While such behavior can be useful for some tasks, it is brittle and will fail as soon as the wind changes. The future will bring humanoids designed to take part in the drama of chaos, inconsistency and error we know fondly as the real world. Such humanoids will not be hindered by complexity and complication, but will embrace it and thrive on it.

Many of the traits we consider uniquely human stem not from our strength, reliability or the precision with which we execute tasks. In fact, we do quite poorly in these arenas. This is not coincidence, but a vestige of our adaptability and ingenuity. Optimality brings stasis and hinders versatility. It is a concept that has little to do with the flux of change in our real world.

As humanoids begin to tap the random variation of the external world and transmute entropy into creativity, philosophers will argue indefinitely about whether they have 'true' intelligence, will, or emotion. Regardless, the majority of us will project these traits into them anyway.

Humanoids will comprise a new distributed infrastructure comprised not only of information, but real-world action. As a given task arises, humanoids will place bids, often partnering with other humanoids to get the job done. Humanoids will not only share workload and resources, but will also evolve by passing host-independent, modular code.

pg. 12

Humanoid Robotics

8. ETHICAL CONSIDERATIONS

The world's population of real humans continues to steadily grow. One might ask why we would want to make a machine that looks, thinks and emotes like a human when we have plenty of humans already, many of whom do not have jobs or good places to live. It is important to reemphasize that humanoids cannot and will not ever replace humans. Computers and humans are good at fundamentally different things. Calculators did not replace mathematicians. They did change drastically the way mathematics was taught. For example, the ability to mentally multiply large numbers, although impressive, is no longer a highly valued human capability. Calculators have not stolen from us part of what it means to be human, but rather, free our minds for more worthy efforts. As humanoids change the contours of our workforce, economy and society, they will not encroach on our sovereignty, but rather enable us to explore and further realize the very aspects of our nature we hold most dear.

So why should we have intelligent, emotion exhibiting humanoids? Emotion is often considered a debilitating, irrational characteristic. Why not keep humanoids, like calculators, merely as useful gadgetry? If we do want humanoids to be truly reliable and useful, they must be able to adapt and develop. Since it is impossible to hard-code high-utility, general-purpose behavior, humanoids must play some role as arbiters of their own development.

Most likely, two distinct species of humanoids will arise: those that respond to and illicit our emotions and those we wish simply to do work, day in and day out, without stirring our feelings. Some ethicists believe this may be a difficult distinction to maintain. On the other hand, many consider ethical concerns regarding robot emotion or intelligence to be moot. According to this line of reasoning, no robot really feels or knows anything that we have not (albeit indirectly) told them to feel or know. From this perspective, it seems unnecessary to give a second thought to our treatment of humanoids. They are not 'real.' They are merely machines.

pg. 13

Humanoid Robotics

The real danger is not that humanoids will make us mad with power, or that humanoids will themselves become super intelligent and take over the world. The consequences of their introduction will be subtler. Inexorably, we will interact more with machines and less with each other. Already, the average American worker spends astonishingly large percentages of his/her life interfacing with machines. Many return home only to log in anew. Human relationships are a lot of trouble, forged from dirty diapers, lost tempers and late nights. Machines, on the other hand, can be turned on and off. Already, many of us prefer to forge and maintain relationships via e-mail, chat rooms and instant messenger rather than in person.

Humanoids are the products of our own minds and hands. Neither we, nor our creations, stand outside the natural world, but rather are an integral part of its unfolding. We have designed humanoids to model and extend aspects of ourselves and, if we fear them, it is because we fear ourselves.

pg. 14

Humanoid Robotics

CONCLUSION

The goal is to create a new kind of tool, fundamentally different from any we have yet seen because it is designed to work with humans as well as for them. Humanoids will interact socially with people in typical, everyday environments. We already have robots to do tedious, repetitive labor for specialized environments and tasks. Instead, humanoids will be designed to act safely alongside humans, extending our capabilities in a wide variety of tasks and environments. Most likely, humanoids will never rise up and wrest control from our hands. Instead, we may give it to them, one home, one factory, one nuclear facility at a time until 'pulling the plug' becomes, at first infeasible and then eventually unthinkable. Even now, imagine the economic havoc if we were to disable the Internet. We are steadily replacing the natural world with the products of our own minds and hands. As we continue to disrupt and manipulate the existing state of our world (often for the better), the changes we make require successive intervention. Technologies engender and demand new technologies. Once unleashed, it is difficult to revoke a technology without incurring profound economic, social and psychological consequences. Rather, the problems that arise from new technologies are often met with more complex and daring technologies.

pg. 15

Humanoid Robotics

BIBLIOGRAPHY

D. Michie, "Machine Learning in the Next Five Years," Proc. of the Third European Working Session on Learning, Glasgow, 1988, pp. 107-122. H.L. Dreyfus, What Computers Can't Do: The Limits of Artificial Intelligence, Harper

Colophon Books, New York, 1979. G.B. Kleindorfer and J.E. Martin, "The Iron Cage, Single Vision, and Newton's Sleep,"

Research in Philosophy and Technology, Vol. 3, 1983, pp.127-142. D.L. Schacter, C.Y.P Chiu and K.N. Ochsner, "Implicit Memory: A Selective Review,"

Annual Review of Neuroscience, Vol. 16, 1993, pp. 159-182. P.E. Agre and D. Chapman, "What are Plans For?" Robotics and Autonomous Systems, Vol. 6, 1990, pp. 17-34. S.F. Giszter, F.A. Mussa-Ivaldi and E. Bizzi, "Convergent force fields organized in the spinal cord," J. Neuroscience, Vol. 13, 1993, pp. 467-491. R. A. Brooks, "Intelligence without Reason," A.I. Memo No. 1293, MIT AI Laboratory, April 1991. T. Inamura, M. Inaba and H. Inoue, "Acquisition of Probabilistic Behavior Decision Model Based on the Interactive Teaching Method," Proc. of the 9th International Conference on Advanced Robotics, Tokyo, 1999, pp. 523-528.

pg. 16