外文文獻(xiàn).pdf

1HUMANOID ROBOTICSHumanoid Robots: A New Kind of ToolBryan Adams, Cynthia Breazeal, Rodney A. Brooks, and Brian Scassellati, MIT Artificial Intelligence LaboratoryASIDE FROM THEIR TRADITIONAL ROLES, HUMANOID ROBOTS CAN BE USED TOEXPLORE THEORIES OF HUMAN INTELLIGENCE. THE AUTHORS DISCUSS THEIRPROJECT AIMED AT DEVELOPING ROBOTS THAT CAN BEHAVE LIKE AND INTERACTWITH HUMANS.In his 1923 play R.U.R.: Rossums Universal Robots, Karel Capek coined robot as a derivative ofthe Czech robota (forced labor). Limited to work too tedious or dangerous for humans, todaysrobots weld parts on assembly lines, inspect nuclear plants, and explore other planets. Generally,robots are still far from achieving their fictional counterparts intelligence and flexibility.Humanoid robotics labs worldwide are working on creating robots that are one step closer toscience fictions androids. Building a humanlike robot is a formidable engineering task requiring acombination of mechanical, electrical, and software engineering; computer architecture; and real-time control. In 1993, we began a project aimed at constructing a humanoid robot for use inexploring theories of human intelligence.1,2In addition to the relevant engineering, computerarchitecture, and real-time-control issues, weve had to address issues particular to integratedsystems: What types of sensors should we use, and how should the robot interpret the data? Howcan the robot act deliberately to achieve a task and remain responsive to the environment? How canthe system adapt to changing conditions and learn new tasks? Each humanoid robotics lab mustaddress many of the same motor-control, perception, and machine-learning problems.The principles behind our methodologyThe real divergence between groups stems from radically different research agendas andunderlying assumptions. At the MIT AI Lab, three basic principles guide our research We design humanoid robots to act autonomously and safely, without human control orsupervision, in natural work environments and to interact with people. We do not design them assolutions for specific robotic needs (as with welding robots on assembly lines). Our goal is tobuild robots that function in many different real-world environments in essentially the sameway. Social robots must be able to detect and understand natural human cuesthe low-level socialconventions that people understand and use everyday, such as head nods or eye contactso thatanyone can interact with them without special training or instruction. They must also be able toemploy those conventions to perform an interactive exchange. The necessity of these abilitiesinfluences the robots control-system design and physical embodiment. Robotics offers a unique tool for testing models drawn from developmental psychology andcognitive science. We hope not only to create robots inspired by biological capabilities, but alsoto help shape and refine our understanding of those capabilities. By applying a theory to a realsystem, we test the hypotheses and can more easily judge them on their content and coverage.2Autonomous robots in a human environmentUnlike industrial robots that operate in a fixed environment on a small range of stimuli, ourrobots must operate flexibly under various environmental conditions and for a wide range of tasks.Because we require the system to operate without human control, we must address research issuessuch as behavior selection and attention. Such autonomy often represents a trade-off betweenperformance on particular tasks and generality in dealing with a broader range of stimuli. However,we believe that building autonomous systems provides robustness and flexibility that task-specificsystems can never achieve.Requiring our robots to operate autonomously in a noisy, cluttered, traffic-filled workspacealongside human counterparts forces us to build systems that can cope with natural-environmentcomplexities. Although these environments are not nearly as hostile as those planetary explorersface, they are also not tailored to the robot. In addition to being safe for human interaction andrecognizing and responding to social cues, our robots must be able to learn from humandemonstration.The implementation of our robots reflects these research principles. For example, Cog (seeFigure 1) began as a 14-degrees-of-freedom (DOF) upper torso with one arm and a rudimentaryvisual system. In this first incarnation, we implemented multimodal behavior systems, such asreaching for a visual target. Now, Cog features two six-DOF arms, a seven-DOF head, three torsojoints, and much richer sensory systems. Each eye has one camera with a narrow field of view forhigh-resolution vision and one with a wide field of view for peripheral vision, giving the robot abinocular, variable-resolution view of its environment. An inertial system lets the robot coordinatemotor responses more reliably. Strain gauges measure the output torque on each arm joint, andpotentiometers measure position. Two microphones provide auditory input, and various limitswitches, pressure sensors, and thermal sensors provide other proprioceptive inputs.Figure 1. Our upper-torso development platform, Cog, has 22 degrees of freedom that we specifically designed to emulatehuman movement as closely as possible.The robot also embodies our principle of safe interaction on two levels. First, we connected themotors on the arms to the joints in series with a torsional spring.3In addition to providing gearboxprotection and eliminating high-frequency collision vibrations, the springs compliance provides aphysical measure of safety for people interacting with the arms. Second, a spring law, in series witha low-gain force control loop, causes each joint to behave as if controlled by a low-frequency springsystem (soft springs and large masses). Such control lets the arms move smoothly from posture toposture with a relatively slow command rate, and lets them deflect out of obstacles way instead ofdangerously forcing through them, allowing safe and natural interaction. (For discussion of Kismet,another robot optimized for human interaction, see Social Constraints on Animate Vision, byCynthia Breazeal and her colleagues, in this issue.)Interacting socially with humansBecause our robots must exist in a human environment, social interaction is an important facet ofour research. Building social skills into our robots provides not only a natural means ofhumanmachine interaction but also a mechanism for bootstrapping more complex behavior.Humans serve both as models the robot can emulate and instructors that help shape the robotsbehavior. Our current work focuses on four social-interaction aspects: an emotional model forregulating social dynamics, shared attention as a means for identifying saliency, acquiring feedbackthrough vocal prosody, and learning through imitation.Regulating social dynamics through an emotional model. One critical component for a sociallyintelligent robot is an emotional model that understands and manipulates its environment. A robotrequires two skills to learn from such a model. First is the ability to acquire social inputto3understand the relevant clues humans provide about their emotional state that can help it understandany given interactions dynamics. Second is the ability to manipulate the environmentto expressits own emotional state in such a way that it can affect social-interaction dynamics. For example, ifthe robot is observing an instructor demonstrating a task, but the instructor is moving too quickly forthe robot to follow, the robot can display a confused expression. The instructor naturally interpretsthis display as a signal to slow down. In this way, the robot can influence the instructions rate andquality. Our current architecture incorporates a motivation model that encompasses these exchangetypes (see Figure 2).Figure 2. A generic control architecture under development for use on two of our humanoid robots. Under each largesystem, we list components that we either have implemented or are developing. Also, many skills reside in the interfacesbetween these modules, such as learning visual-motor skills and regulating attention preferences based on motivationalstate. We do not list machine learning techniquesan integral part of these individual systemsindividually here.Identifying saliency through shared attention. Another important requirement for a robot toparticipate in social situations is to understand the basics of shared attention as expressed by gazedirection, pointing, and other gestures. One difficulty in enabling a machine to learn from aninstructor is ensuring the machine and instructor both attend to the same object to understand wherenew information should be applied. In other words, the student must know which scene parts arerelevant to the lesson at hand. Human students use various social cues from the instructor fordirecting their attention; linguistic determiners (such as this or that ), gestural cues (such aspointing or eye direction), and postural cues (such as proximity) can all direct attention to specificobjects and resolve this problem. We are implementing systems that can recognize the social cuesthat relate to shared attention and that can respond appropriately based on the social context.Acquiring feedback through speech prosody. Participating in vocal exchange is important formany social interactions. Other robotic auditory systems have focused on recognition of a smallhardwired command vocabulary. Our research has focused on understanding vocal patterns morefundamentally. We are implementing an auditory system to let our robots recognize vocalaffirmation, prohibition, and attentional bids. By doing so, the robot will obtain natural socialfeedback on which actions it has and has not executed successfully. Prosodic speech patterns(including pitch, tempo, and vocal tone) might be universal; infants have demonstrated the ability torecognize praise, prohibition, and attentional bids even in unfamiliar languages.Learning through imitation. Humans acquire new skills and new goals through imitation.Imitation can also be a natural mechanism for a robot to acquire new skills and goals.4Consider thisexample:The robot is observing a person opening a glass jar. The person approaches the robot and placesthe jar on a table near the robot. The person rubs his hands together and then sets himself toremoving the lid from the jar. He grasps the glass jar in one hand and the lid in the other and beginsto unscrew the lid by turning it counter-clockwise. While he is opening the jar, he pauses to wipe hisbrow, and glances at the robot to see what it is doing. He then resumes opening the jar. The robotthen attempts to imitate the action.Although classical machine learning addresses some issues this situation raises, building asystem that can learn from this type of interaction requires a focus on additional research questions.Which parts of the action to be imitated are important (such as turning the lid counter-clockwise),and which arent (such as wiping your brow)? Once the action has been performed, how does therobot evaluate the performance? How can the robot abstract the knowledge gained from thisexperience and apply it to a similar situation? These questions require knowledge about not only thephysical but also the social environment.4Constructing and testing human-intelligence theoriesIn our research, not only do we draw inspiration from biological models for our mechanicaldesigns and software architectures, we also attempt to use our implementations of these models totest and validate the original hypotheses. Just as computer simulations of neural nets have been usedto explore and refine models from neuroscience, we can use humanoid robots to investigate andvalidate models from cognitive science and behavioral science. We have used the following fourexamples of biological models in our research.Development of reaching and grasping. Infants pass through a sequence of stages in learninghand-eye coordination.5We have implemented a system for reaching to a visual target that followsthis biological model.6Unlike standard kinematic manipulation techniques, this system iscompletely self-trained and uses no fixed model of either the robot or the environment.Similar to the progression observed in infants, we first trained Cog to orient visually to aninteresting object. The robot moved its eyes to acquire the target and then oriented its head and neckto face the target. We then trained the robot to reach for the target by interpolating between a set ofpostural primitives that mimic the responses of spinal neurons identified in frogs and rats.7After afew hours of unsupervised training, the robot executed an effective reach to the visual target.Several interesting outcomes resulted from this implementation. From a computer scienceperspective, the two-step training process was computationally simpler. Rather than attempting tomap the visual-stimulus locations two dimensions to the nine DOF necessary to orient and reach foran object, the training focused on learning two simpler mappings that could be chained together toproduce the desired behavior. Furthermore, Cog learned the second mapping (between eye positionand the postural primitives) without supervision. This was possible because the mapping betweenstimulus location and eye position provided a reliable error signal (Figure 3). From a biologicalstandpoint, this implementation uncovered a limitation in the postural primitive theory. Although themodel described how to interpolate between postures in the initial workspace, no mechanism forextrapolating to postures outside the initial workspace existed.Figure 3. Reaching to a visual target. Once the robot has oriented to a stimulus, a ballistic mapping computes the armcommands necessary to reach for that stimulus. The robot observes its own arms motion. It then uses the same mappingthat it uses for orientation to produce an error signal it can use to train the ballistic map.Rhythmic movements. Kiyotoshi Matsuoka8describes a model of spinal cord neurons that producerhythmic motion. We have implemented this model to generate repetitive arm motions, such asturning a crank.9Two simulated neurons with mutually inhibitory connections drive each arm joint,as Figure 4 shows. The oscillators take proprioceptive input from the joint and continuouslymodulate the equilibrium point of that joints virtual spring. The interaction of the oscillatordynamics at each joint and the arms physical dynamics determines the overall arm motion.Figure 4. Neural oscillators. The oscillators attached to each joint comprise a pair of mutually inhibiting neurons. Blackcircles represent inhibitory connections; open white circles are excitatory. The final output is a linear combination of theneurons individual outputs.This implementation validated Matsuokas model on various real-world tasks and provided someengineering benefits. First, the oscillators require no kinematic model of the arm or dynamic modelof the system. No a priori knowledge was required about either the arm or the environment. Second,the oscillators were able to tune to a wide task range, such as turning a crank, playing with a Slinky,sawing a wood block, and swinging a pendulum, all without any change in the control systemconfiguration. Third, the system was extremely tolerant to perturbation. Not only could we stop andstart it with a very short transient period (usually less than one cycle), but we could also attach largemasses to the arm and the system would quickly compensate for the change. Finally, the input to the5oscillators could come from other modalities. One example was using an auditory input that let therobot drum along with a human drummer.Visual search and attention. We have implemented Jeremy Wolfes model of human visual searchand attention,10combining low-level feature detectors for visual motion, innate perceptualclassifiers (such as face detectors), color saliency, and depth segmentation with a motivational andbehavioral model (see Figure 5). This attention system lets the robot selectively direct computationalresources and exploratory behaviors toward objects in the environment that have inherent orcontextual saliency.Figure 5. Attention system overview. Various visual-feature detectors (color, motion, and face detectors) combine with ahabituation function to produce an attention activation map. The attention process influences eye control and the robotsinternal motivational and behavioral state, which in turn influence the weighted feature-map combination. We captured theimages during a behavioral trial session.This implementation has let us demonstrate preferential looking based both on top-down taskconstraints and opportunistic use of low-level features.11For example, if the robot is searching forocial contact, the motivation system increases the weight of the face-detector feature. This producesa preference for looking at faces. However, if a very interesting nonface object appeared, theobjects low-level properties would be sufficient to attract the robots attention. We areincorporating saliency cues based on the models focus of attention into this attention model. Wewere also able to devise a simple mechanism for incorporating habituation effects into Wolfesmodel. By treating time-decayed Gaussian fields as an additional low-level feature, the robot willhabituate to stimuli that are currently receiving attentional resources.Shared attention and theory of mind. One critical milestone in a childs development is therecognition that others have beliefs, desires, and perceptions independent of the childs. The abilitiesto recognize what another person can see, know that another person maintains a false belief, andrecognize that another person likes games differing from those the child enjoys are all part of thisdevelopmental chain. Furthermore, the ability to recognize yo