Service robots are a relatively new field that has been trying to establish a market. ASIMO is a humanoid robot that was designed to operate in a human living space. The biped robot was created by Honda to drive innovation within the organization and to assist humans with their daily tasks. The focus of this paper is to highlight the internal and external elements that shaped the systems architecture of ASIMO in order to give it human resemblance as well as establish a path towards usefulness. These include engineering challenges as well as social, economic and political factors.
Introduction
Since early times humans have tinkered with the idea of creating machines that can replicate human tasks. From Leonardo Da Vinci’s Mechanical Knight1 to modern works of fiction, engineers have pushed the boundaries of creativity to bring robots to life such as in factory floors. The advance in computers has enabled engineers to create machines with highly computer-intensive actions such as mobility, perception and vision. These systems have started to gain anthropomorphic characteristics as well as intelligent behavior with the vision coexisting and assisting humans. They exist in this form to better communicate with humans and adapt to other systems that were built with the human form. What started as an iterative process, has now amassed many components and subsystems with increasing complexity. One of these systems is ASIMO, a humanoid assistant robot develop by Honda’s R&D. As the robot will be interacting and assisting humans, safety and trust are crucial. Therefore it is important to architect humanoid robot systems in a manageable method in order to contain complexity and reduce emergent behavior. External influences such as social, economic and political aspects also play an important role in the overall architecture of the system.
The focus of this paper is to highlight the internal and external elements that shaped the systems architecture of ASIMO in order to give it human resemblance as well as establish a path towards usefulness. The observations mentioned by the author are intended to recreate the system from a systems architecture perspective and may not represent the actual process taken by the engineers. The use of heuristics has been incorporated as evocative guides to highlight important aspects that went into the architecture. They recreate lessons to be followed but not to be quoted from the architect(s) directly. Throughout this paper the word ASIMO, robot and system are interchanged to recreate different viewpoints. The mention of Honda encompasses the larger system view and wider array of indirect factors towards the program, while the mention of architect(s) encompasses a more limited and direct view of the program.
Background
In the 1970’s Honda Motor Co. was struggling to catch up with competitors in the auto industry and was in search of new ideas. Honda soon realized that electronics held the key towards evolution and differentiation in the business. In 1986, Honda created a research division to drive innovation in the areas of mobility. Several projects in different areas were being pursued and one of them was the creation of a humanoid robot. Masato Hirose was appointed to lead and create a robot that resembled the “Mighty Atom2”. With no potential business, market or prior experience in robots, Hirose embarked in the mission to create a two-legged-robot that could walk like a human.
Complex systems will develop and evolve within an overall architecture much more rapidly if there are stable intermediate forms than if they are not [9]
There were no specific requirements provided to Hirose other than to build a human robot that could walk; the rest of the details where left to his imagination and research. For the first years Hirose and his team studied biological systems to understand how bipedal locomotion system worked. Using the models they developed by observing animals and humans walking patterns, the first static walking prototype known as the “E0” was developed in 1986. This was essentially a pair of legs that could recreate human steps, albeit slow and stiff. In the following years better, faster and dexterous prototypes were created as part of the E Series. These prototypes where complex at their time because of the lack of understanding of walking behavior and the system could not have evolved without building it at intermediate stages. Once a concept was proven, a new prototype was built from scratch and the architecture incorporated new technology, sensors, joints and control software. When the basics of walking, stability and balance were achieved, the next phase of the system incorporated torso, arms, and ahead to extend functionality, known as the P series. By 1993, the system was a 1.9 m (~6 ft) tall heavy weight robot that possessed great force. It could grab objects and change their position. Figure 1 shows the time line of the system’s evolution starting from the first prototype E0 to the modern ASIMO.
Figure 1. Evolutionary time lapse of the system
As the robot started adopting more human-like features, Honda even went to the extent as to consult with the Vatican, which came to terms that humanoid robots could exist for human good [1]. The next goal was to give the robot anthropomorphic characteristics and behavior. The robot has to portray a friendly design, similar to how a car’s design and usability attracts a customer. Additionally, interaction is important for any kind of robot to perform tasks in human and has to understand human's high-level task requirements by using its low-level sensory modules such as eyes, ears and tactile sensors [13].
In 2000 Honda announced ASIMO (an acronym for Advanced Step in MObility), a humanoid robot with 34 Degrees of Freedom3 (DOF) that is 1.3 m (~4 ft) tall, weights 54kg (120 lbs) and can run up to 6 km (3.7 miles) per hour [6]. This was the product of research and studies that lead to the ideal size and shape. ASIMO was conceived to function in an actual human living environment in the near future. It is easy to operate, has a convenient size and weight and can move freely within the human living environment, all with a people-friendly design [3]. New ASIMO (2005) is an updated version that holds similar resemblance to the first ASIMO but with added social & mobility capabilities.
System Description
The greatest leverage in system architecting is at the interfaces [9]
A high level representation of the system is shown in Figure 2. The system can be broken into 8 elements: Main Control, Arm Leg Head Actuators, Communication, Voice, Vision, Navigation, Sensors and Power Control. These elements are represented within the dotted box. They are physically connected and exchange power and information. The items outside the systems are elements on which the system interacts or influence the system. These can be simple objects, people, terrain or sounds in the environment. They are mainly captured by the system via cameras and microphones. There are other sensors placed in different parts of the structure to provided additional inputs on the location, position and distance of its surroundings. Though the system has some level of autonomy, it still requires intervention by an operator for more difficult tasks such as thinking. The communication module maintains a wireless link with the dedicated ASIMO service. This allows the operator to have a direct connection with the system. The autonomous parts of the system can also directly query data from the server to make decisions.
Figure 2. High level system decomposition and interfaces. Expanded from [5]
As a source of power, a battery is attached to ASIMO’s back, similar to an astronaut’s backpack. This provides the system with all the power necessary for a few hours to move around without the need of a physical cable. When combined together, the system harmoniously reacts to give the structure a human like movement. Through visual and sensory inputs, the system can compute a path or a mission and instruct the joints to coordinate to reach or achieve a task. Each of the major elements can be considered as separate systems with their own degree of complexity. Keeping the elements well defined can help contain complexity as functionality grows. The goal is to keep the organization simple and keep the flow of information between the interfaces to a minimum.
The following sections describe some of the important architectural elements of the system and their relation to biological systems.
Mobility
One of the most remarkable characteristics of ASIMO is the ability to walk like a human. With 6 DOF on each leg, it can attain bipedal motions such as walking, running, shifting from side to side, rotate in place, kick a ball and climb stairs to name a few. The system not only computes and balances its own weight, but also compensates for objects being carried or pushed against its own weight. The challenges of stable walking are [6]:
- Not falling down even if the floor is uneven
- Not falling down even when pushed
- Being able to walk stable on stairs or slopes
The human skeleton was used for reference when locating the leg joints [5]. Initially, finding the perfect balance between mobility and flexibility took several iterations by building a prototype and observation the reaction. When the robot fell over, the team carefully reviewed what happened and reprogrammed the mechanism and it was a repetition of trial and error [17]. Once the prototype was stable at a certain speed, Hirose’s team would increase the speed and start over if the robot fell again.
The movement of a robot causes acceleration and deceleration in all parts of its body, which generates inertial forces. The robot is also subject to the force of gravity. The cumulative effect of all forces is termed the total inertial force [23]. Joint rotations are precisely orchestrated to achieve overall motion. A successful movement transition will result in one step. The speed and direction of each steps group together will define the action such as walking, running or rotating. These actions will be driven by an objective which is determined by a computation in the navigation element. The result of the computation is a sequence of footstep placements that reach a goal region while minimizing encoded heuristics for effort, risk, or the number and complexity of the steps taken [10]. As new technology became available, the team moved to computer simulation to model walking to analyze the behavior without having to build a prototype. Figure 3 shows how the robot attains balance by using its upper body as well as the force exerted when moving. The faster the legs move (i.e. running), there will be a small duration in which both legs will be in the air. One important aspect of the design involved reducing the shock when a leg lands on the ground. By modeling the movement and force of the leg, the team was able to add shock absorbers in the sole of the foot to improve stability.
Figure 3. Balancing motion and foot landing force simulation [32]
The overall requirement is that the robot should be able to operate in the same environment as humans. Such environments can have different obstacles that require a different set of actions to navigate through them. Legged robots have the unique ability to step onto or over obstacles or unsafe footholds, which can allow them to traverse terrains that would be impassable to a wheeled mobile robot [10]. This makes bipedal robots a better choice, but more complex, when defining the mobility method of the system in the architecture.
Manipulation
There are 23 muscles in a human arm and about 40 tendons and 20 muscles in the hand [22]. A complex system of its own, the human arm and hand act as tool to achieve tasks that require precision and dexterity. Robotic arms have been widely used in industrial settings for several decades and have achieved high precision and strength. However, the same cannot be said about robotic hands since the tip of the industrial robotic arm is usually equipped with a specialized tool. In order to design and model the same concept to robotics, the human hand has been subject of study. In particular joints and bones can be replaced by revolute or universal joints and links are considered as rigid bodies [23].
Figure 4. ASIMO opening a flask
Hirose’s aim was to have a robot that could go get a can of beer from the refrigerator [17]. There were two challenges, one being how to open the door of the refrigerator and the other how to actually navigate internally and grab the beer. Though the task was simple, recreating the task via a robot has proven to be more complicated than initially thought. ASIMO’s arm and hand have 9 DOF not including fingers. Combined together, these allow the robot to reach for and grab for a wide variety of objects within proximity. Tactile sensors are embedded in the fingertips. These enable the robot to remove the lid from a flask and pour beverage into a cup, use sign language [11] as well as hold onto objects and perform gestures (body language). The hands and arms rely extensively on the inputs from the Vision System prior to any task execution for object identification, points of grasp, type of task and direction.
One could argue that there are more practical forms (claws or suction) of achieving such tasks, after all humans have developed a wide range of tools to augment their own hands. However, this would change the structure of the robot and the goal is to keep the structure robot as human-like as possible. This is not only done for the comfort of the human user but for the multi-functional uses it can have. A robot hand, while still in a primitive form, can keep an “open architecture” at the interface. This is especially important in new markets in order to keep all options open once the market starts to respond [12].
Perception
Humans naturally perceive their environment through the visible spectrum (390 to 700 nm wavelengths). With the aid of computers, we have evolved to augment the perception of the environment by tapping into other spectrums. Rechtin (1991) refers to the use of “false color” by the means in which machines allow us to heighten both brightness and color contrast through the visible range. We can “see” beyond visible wavelengths and can visualize other parameters (such as heights, structural stress, molecular presence, etc) as high-contrast patterns in color and space. Therefore, when referring to vision, we can include all visible and non-visible elements to humans, as long as there is a tool that allows such element to be perceived.
Figure 5. Distribution of sensors
When considering vision in a robot, human visible and non-visible elements can also be considered. This can primarily be attributed to the fact that cameras have not reached the same level of perception and understanding of the human eye, which leaves us to the use of other aids. ASIMO is equipped with two cameras to read contextual data, a ground sensor to scan ground surfaces, and an ultrasonic sensor to detect obstacles that cannot be seen (e.g. glass). The vision and navigation system interact together to extract 3D and 2D objects [13]. The robot then uses this data for recognition of moving objects, human postures and gestures, and face recognition [6]. Figure 5 shows where each sensor is located in the robot and the range of perception.
The field of AI is still far from optimal and is one of the major constraints keeping ASIMO limited in functionality. In some situations it could be easier and necessary to incorporate multiple sensors such as radiation, temperature or air quality into the architecture to complement the visual perception. In others, there is no known sensor that can substitute contextual data and the only form of detection is through analysis of the visual domain. For example, how to distinguish if fire inside a house is safe? Should it be extinguished if it’s originating from the stove? As the domain of computer vision continues to evolve, ASIMO should be able to recognize more objects as well as situations.
Communication
Language is the primary mean in which humans communicate with other humans. It is an abstract concept that can be shaped on a number of factors and can be used to convey a circumstance or request. Rechtin (1991) defines languages as a system that:
- Have different forms for different functions
- Are usually unbounded
- Are structured according to rules
- Contain models or perceptions of the world around them
- Are adapted, advanced, or removed, as circumstances dictate
ASIMO can speak in several languages and incorporate gestures in the conversation. It can also recognize a variety of sounds, gestures and symbols (encoded). It uses a set of microphones to capture sound for further processing of the signals. Voice activity detection (VAD) and sound source separation are essential for robots to communicate with people in real environments [14]. ASIMO also relies on visual inputs to lock the source or subject that is generating the request. Additionally, it converts the gestures into sounds and vice versa. For example, pointing to an object or direction can result in a verbal confirmation. Saying “goodbye” can result in ASIMO waving his hand.
As with other computers, ASIMO can also communicate with other computers, and much more efficiently than with humans. There is much more processing needed to decode human language than computer protocols. Computer requests and responses are designed to be logical and direct interpretation. ASIMO communicates wireless with an external server to retrieve and store information.
Rechtin (1991) proposes that the best man-machine communication is a bridging language that both humans and machines can learn, understand, and use with the least total difficulty [9]. This intermediate form of communication is being used with IC communication cards, an IC tag with optical communication functions, so that ASIMO can autonomously select and execute tasks when a card is within proximity [6]. Users and guests can use this tool as an alternate communication method which can command the robot without using voice or gestures.
Thought
ASIMO is far from achieving the same level of thinking as a human. That would require having knowledge on how the human brain works, and what little is known, is that the human brain’s architecture is not fixed and preprogrammed [9]. Computer thinking is constructed the opposite way, and is mostly driven by the amount of data that can be computed in a given time.
ASIMO’s intelligence can be evaluated by the number of autonomous actions it can perform and the amount of information that is retained or accessed. The system architecture is behavior-based which combine deliberative planning like high-level human commands and reactive behavior control for navigation in a rapidly changing environment [13]. Reactive behaviors are mostly controlled internally by the system. Deliberate behaviors require the assistant of an external system. This holds a database system to retrieve individual information (biography and biometrics), a reservation system to administer corporate events, and a global map system to plan navigation and interaction. Additionally, it provides a gateway to the internet where information such as weather, news or other topics of interest can be retrieved.
ASIMO Architectural Consideration
Intelligent System Behavior
Different architectures can generate different behavior [9]
The elements discussed in the previous sections are common properties of humans, a biological system. Each element can act as standalone systems and have different architectures with complexities and challenges within their respective domains (software and hardware). When combined together, the overall architecture changes to achieve a greater purpose which resembles that of an intelligent system.
Figure 6 shows the behavior elements and their relations that intelligent systems share in common. The system will reside in a given space and time (environment) which influences its purpose. ASIMO is designed to operate in the same habitat as humans, mainly the household and office space. It will encounter humans, daily objects and obstacles in this space. The system therefore needs inputs (sensing) and outputs (actions) to interact with its environment. ASIMO uses vision and other sensors to sense its environment and uses its legs and arms to perform an action. The system needs to have a brain (thinking) that will process external as well as internal data and convert them to action in order to fulfill a purpose. ASIMO’s brain allows it to walk and interact with its environment as well as monitor its own health. The system will normally need to interact (communicate) with other systems to exchange information and grow. ASIMO uses human language to interact with humans as well as computer protocols to fetch knowledge from cyberspace. Future research will allow it one day to collaborate with another ASIMO to achieve broader set of tasks. What one ASIMO learns can be uploaded to the external system and shared with a different ASIMO to create collective learning.
The initial system would lack enough intelligence to make complicated decisions so it has to rely on preprogrammed motions. Hirose’s initial approach was to model the relationship between sensory information and behavior directly [13]. This could be compared to a muscle reflex, such as when your hand touches something hot and instinctively move your hand away. The biological system sends various signal-and-image-processing centers send their results to several other locations for further processing or action and the quickest results go to the motor centers for reflex control [9].
Figure 6. Intelligent Behavior System decomposition [5]
How intelligent should the system be? Hirose views robots as simply an extension of a machine, so they should be a tool that is convenient to use without the ability to makes its own judgments [19]. The intelligence level referred in this section is trying to achieve very basic activities. However, what appears as a simple and normal task to others humans are in reality quite complex to implement. Due to the growing nature of the system, the system should be designed to support many complex functionalities but easy to adapt and comprehend. A strategic choice taken by the team was to work on basic autonomous behaviors such as balancing and obstacle avoidance within the system, and leave other higher intelligence tasks (non real time and slow) to the operator (outsourcing the brain). By making the system partially manned, the system was allowed to progress as a whole without investing too many resources on the intelligent aspect, with the goal to moving to moving to a fully automated as technology improved.
Except for good and sufficient reasons, functional and physical structures should match [9]
The software architecture has been elegantly structured to correlate with the functionality of the physical aspect. This may be influenced by the fact that the research team was initially composed of mostly mechanical engineers. Therefore, most of the structuring and software characteristics may have been modeled after tangible objects. Figure 7 illustrates the hardware and software structure. In the vertical axis, the system is grouped by functionality and has a layered design. In the bottom lies the mechanical control components, these are the “acting” behaviors that interact with the environment in form of physical actions. These give the robot mobility to reach a destination (e.g. go to the kitchen) and ability to manipulate the position and form of objects (e.g. pour orange juice in a cup). Next up are Vision & Auditory components which act as sensory and communication devices. These provide the robot with information of its surroundings as well exchange information. Planning components follow, which give the system local thinking capabilities to make autonomous decisions. The thinking capability is aided by External System & Resources components. In a layered system, the elements of a layer interact to produce a set of services, which are made available to next higher layer [12]. The physical layer objects lie in the bottom and interact directly with the hardware. On the very top lie the applications that drive the system to achieve a greater purpose, in other words the services that give the system intelligence. As technology evolves, some of the services offered in the external system could be moved to the system itself. Likewise, if physical components such as hands get replaced, the top layers should not be affected other expanding the tasks the robot is capable of performing. Having a layered design makes the architecture easily adaptable to these types of changes, contains complexity within the element and keeps the architecture manageable.
In the horizontal axis, the system is grouped by physical parts. On the left are the legs, which are responsible for the mobility of the system. In the middle are the arms, legs & torso that provide interaction with the environment. In the right is the head contains the sensory inputs and outputs. The diagram acts as a map of the structure, by specifying the functionality and component location of interest the subsystem can be identified. When relating activities, the reaction from one activity resonates throughout the system. For example, while the robot is carrying out a mission such as walking from point A to B, it utilizes visual inputs to make a virtual map of its environment. 3D models are computed to detect the terrain (slope, stairs, etc) as well as people. At the same time, moving objects have to be mapped and predict their paths in order to avoid collision. If the object happens to be a person, the individual data is extracted by doing face recognition. All this information loops as the robot advances and queries the server for supplemental information.
Figure 7. High level function and software
Anthropomorphism
The overall goal of ASIMO is to interact and blend in the same environment as humans. Honda wanted to achieve this goal by giving the robot an anthropomorphic design. But why go through such extent when the same functionality can be achieved in more practical forms? Humans have built tools, vehicles and structures around the human shape. From spoons to cars and skyscrapers, they are all designed to be operated and inhabited by humans. In other words, systems should be designed to serve the needs of humans. Since ASIMO is designed to assist humans, it should have the capability of navigating and using the same artificial obstacles and tools found in a human environment (brooms, doors, stairs, roofs, etc). The best way to research and develop an intelligent system with human like intelligence is to give similar traits and representation.
In introducing technological and social change, how you do it is often more important than what you do [9]
Human psychology and perception also plays a role. The earlier designs of the robot focused on achieving mobility, balance and efficiency. The first humanoid robot prototype was a 1.9 m (~6 ft) tall, 175 Kg (386 lbs) robot known as the P1; recall Figure 1. While the functional goal had been somewhat achieved, the P1 was far from looking friendly and didn’t quite fit the environment it was being designed for. For the next models, Hirose and his team focused on refining the physical shape and appearance of the robot to better blend with humans. The P2 and P3 hid the mechanical aspect, reduced height & weight, and incorporated more autonomous functions. For ASIMO, Hirose’s team focused on simplicity and human qualities as a foundation to create mockups and simulations. Several prototypes of several sizes, proportions and forms of the robot were created to explore this domain. Computer rendering and simulation was then used to validate the constraints of the system, especially how mobility would be affected.
Lacking customer opinion, Hirose’s team relied on user and public input to style ASIMO. Image surveys were conducted concerning ways it would be used in the market, and styling designs clinics were also conducted to hear views inside and outside Honda. Further evaluation of the model was done by presenting both real and digital versions of the robot on TV and special events [15]. With enough feedback on the design and additional verification, the next iteration of the system took place. A strategic choice was to give ASIMO the appearance of a 10 year old boy. The robot’s short height invites even a child to exert dominance over it, while being tall enough to operate objects in a household or office. In other words, it’s not enough to just achieve a specific engineering goal (human-like mobility) but execute it in an aesthetical pleasing way.
How human should the robot be? Studies have predicted that too much perceived similarity between social robots and humans triggers concerns about the negative impact of this technology on humans and their identity more generally because similarity blurs category boundaries undermine human uniqueness [24]. Therefore the robot must still posses certain characteristics that distinguish human from robot. Additionally, there is a stigma that machines will replace and even exterminate humans. But it is also undeniable that automation has also introduced many positive effects into society. Adding human resemblance and emotional value masks the inner difference of a machine. It becomes easier to accept that a friendly ASIMO can be left at home tending to basic needs of your house. The architect therefore must add the human touch to the system, a complex behavior not yet well understood.
A Disruptive System in the Making
Systems architecting is a process driven by a client’s purpose or purposes [12]. In this case, Hirose’s direct client was Honda’s top management, and had no external customer influence for several years due to the program’s secrecy. Architecting a system that caters to the organization’s own vision with hopes of future commercialization can be a challenging task. Therefore Hirose’s team needs to foresee the type of demands and necessities that can be addressed. This requires a set of knowledge areas that span beyond technology and can go well into demographic, political, and market trends to name a few. Even if the system does happen to capture a market, it may not capture the entire needs of customers. The system architecture will need to change as well, but doing so minimally, can accelerate the system into production without much delay.
In the case of disruptive technologies, sometimes it’s necessary to first develop the product in order to create demand for it. Early architects4 such as Henry Ford once said “if I had asked people what they wanted, they would have said faster horses.” Even in modern era, the late Steve Jobs said “customers don't know what they want until we've shown them.” They both created disruptive systems that identified a latent need and presented it to the public. Likewise, consumers may be asked what would solve their problems at home or at work and their solution may be within a limited reality, and not a robot, especially one with humanoid features. Hirose’s team and Honda’s management are taking a greater challenge and risk since the completion of the system will take many years, and will need technology that is still not available. Perhaps the customers of this system are the future generations to come.
Unprecedented systems have been both purpose driven and technology driven [12]. On the one hand, Honda wants to push the boundaries of imagination to create disruptive technology that will improve its products and services. On the other, it wants a humanoid robot that can coexist with humans and be of assistance. Figure 8 shows how Honda gradually seeks to introduce ASIMO in limited domains and slowly expand its use as the technology improves. Once the technology is useful for humans, it can aggressively start seeking wider markets.
Figure 8. Development of ASIMO roadmap [32]
In its current form, the system merely offers an entertainment service, at most an expensive computer assistant that could easily be replaced by a lower cost human. Nonetheless, as the system advances it has the potential to compete in the market. Factors such as a shift in population, increasing wages and decrease cost of technology with higher capability can contribute in making the system adoptable by business and households over time. Honda foresees that the system to evolve to the point where it can be adopted by customers. Considering the aforementioned factors and continuous funding, the system is well posed to reach the usability threshold. For example, dataprocessing speed and software improvements allow the commercial version of ASIMO to walk up or down stairs without preprogramming. The earlier version needed to be programmed in advance with the data (width, height, etc.) about the specific stairs [33]. Other system functionalities are also bound to see improvements with time.
Long term programs also need the financial backing and a work culture that embrace bold moves. For instance, Honda’s automobile and motorcycle businesses are the driving force behind its finances and methodologies. Every CEO in Honda’s history has had an engineering background which has kept the company biased towards an indispensable aspect of the engineer craft [34]. Making funding research programs like ASIMO have strong support from upper management, even if they seem out of the ordinary. This along other factors drive to what is known as the “Honda Effect”, in which strategy is not a long term concrete business plan but rather the things necessary for the successful functioning of an organizations an adaptive mechanism [34].
Production and Metrics
ASIMO is available for commercial use and is being used receptionist work as well as for demonstration purposes. The estimate cost of an ASIMO is approximately \(1 million5; a unit can be instead leased for \)150,000 a month [25]. The cost is mainly due to the limited robots produced (in the double digits) and because they are produced and assembled by hand. Among the initial adopters of the system in Japan are museums and IBM. In America, Disneyland park guests could see ASIMO interact with a live host and operate in a modern home environment by using all of its unique technological capabilities, such as walking forward and backward, climbing and descending a flight of stairs and taking direction from a person [26]. The exhibition was sponsored by Honda and there are no official figures released on the estimated revenue or how many visitors the show attracted. Among these users, the author is currently aware of their remaining use in museums such as a Tokyo’s National Museum of Emerging Sciences and innovation Miraikan. Whether these exhibitions are currently generating revenue for Honda has not been disclosed. Additionally, there are a few obstacles other than the price tag. The batteries only last a few hours; and the more intensive the physical task, the faster power will be consumed. Honda will need to increase the duration of the battery in order for the system to achieve significance tasks. Also, the AI of the system is very limited and requires operators and staff to support ASIMO. The system should be autonomous and require minimal support in the consumer market. Once these hurdles are overcome, Honda can use its leverage to manufacture the systems using its existing line of industrial robots at a fraction of the cost. Honda’s goal is to eventually sell ASIMOs to consumers for about the cost of a car [19].
The most important lesson in revolutionary systems, at least those not inextricably tied to a single mission, is that success is commonly not found where the original concept thought to be [12]
Success can be measured in multiple ways. From a commercial perspective, the system hasn’t been fully launched into the market and it is unknown whether ASIMO generates a profit for Honda or have any demand at all. From organizational perspective, the system has been an engineering success as it achieved one of its goals: generate new ideas that can create new opportunities for the company. Some of its technologies have transitioned from humanoid to its automobile business and other emerging areas. Honda’s automotive engineers borrowed visualization technology from the humanoid program for the company’s Lane Keeping Assist System, a safety system in which a camera monitors lane lines, allowing steering assistance to keep the car centered [16]. While this technology might not have been considered back in 1986 in Honda’s automobile roadmap, the humanoid robot project had unintentionally set in motion the technology to be conceived early on. Computer Vision is now being aggressively perused by other automakers and technology companies (such as Google and Apple) to achieve self-driving cars. Other robotic products that derived from ASIMO include Walking Assist device with stride management system which supports walking for people with weakened leg muscles, and the Bodyweight Support System which reduces the load on legs by supporting part of the bodyweight [28]. These products aim to assist people with disabilities to walk and to augment human strength. While they haven’t been declared as an imminent success, these are the type of unexpected consequences that Honda hoped would emerge by pushing technology and imagination to its limits in the robotics division. If not useful in its initial form, the technology will continue to evolve and find an application in some other form.
Public demonstration of the system has also generated free marketing for Honda which has brought public awareness of Honda’s brand beyond cars and motorcycles. ASIMO has been established as an icon that represents Honda as a research power house.
Competition, Politics and Cultural Factors
Predicting the future may be impossible, but ignoring it is irresponsible [12]
The world is headed towards automation and some companies are investing in the research and development of humanoid robots hoping to catch the next revolutionary wave. While industrial robots are a multi-billion market, the same cannot be said of service robots. A number of other Japanese companies, including NEC, Sony, Hitachi ltd and Mitsubishi, have developed robots with humanoid features or functions, but so far none has caught on in a big way with consumers [27]. The political and cultural setting in Japan may be some of the reasons why companies feel compelled to participate. In 1998, the Ministry of Economy had driven similar efforts by funding the Humanoid Robotics Project for 5 years [4]. In 2014, Prime Minister Shinzo Abe described plans to modernize Japanese economy by investing in robotics [27]. This has established a fertile ground where companies have setup several research initiatives. Though most will not survive, having a competitive setting is what keeps pushing companies and institutions to develop better solutions.
A recent contender is Softbank, a Japanese telecommunications company, which introduced a 1.2 m (4 ft) tall semi-humanoid robot called Pepper to the market last year. Whereas Honda has focused on a robot that can assist humans in everyday tasks, Softbank is targeting emotional robots that can relate to human feelings. Interestingly enough, the strategy involves a more aggressive business stance to fully utilize the current technology and bring service robots to mainstream. By acquiring Aldebaran Robotics6 in 2013 and partnering with Foxconn7 and Alibaba8 and, Softbank leveraged technology, manufacturing and distribution channel to commercialize humanoid robot. At about \(10,000 for a 36 month subscription, Softbank is selling the hardware at a loss to break the expensive bias that comes with robots, and instead recoup by selling software as a service. The business model is similar to how wireless carriers partially subsidize the cost of a smart phone in exchange for a higher monthly wireless plan.
Softbank also accelerated product commercialization by removing the complexity involved with bipedal mobility. Instead, Pepper is conformed of a single piece from the hips down and uses wheels to move. The robot recognizes speech and facial expressions and uses its arm for body language. A tablet is attached to its chest for supplementary visual and tactile interaction. The robot is intended to play a variety of roles, from nurse to babysitter to portable entertainment system provider [27]. The system can be added “intelligence” by downloading applications from their service cloud. The intention is to create an “app” (application) environment such as Apple and Android to supplement revenue. In regards to emotional robots, Hirose argues that when developing robots as useful machines, you cannot disregard their interaction with human beings [17]. Just as some drivers develop an attachment to their cars when driving them, Hirose intends to create a similar attachment to robots when using them.
In the U.S., there appears to be a stigma towards robots that have a human appearance. Perhaps the strong predominance of movies depicting robots taking over the world (as life and jobs) has stroked awe and fear among the public. Hirose’s team dedicated a lot of time in giving ASIMO a “friendly” appearance, however President Barack Obama first impression of ASIMO was describe as “a little scary” [31] when interacting with the robot during one of his visits to Japan. Though there was recognition of the innovation and technology, the perception conveyed by such statement can influence the private sector in delaying the adoption of these systems.
Another example is American company Boston Dynamics9 which had been acquired in 2013 by Google. After losing its primary intended customer, the U.S. military, it proceeded to enter the humanoid robot arena. After demonstrating to the world their 1.88 m (~6 ft) tall humanoid robot named Atlas with capabilities to lift heavy boxes, relationship with Google went sour. Perhaps it was its mechanical appearance, the perception that its ability to carry boxes would displace workers, or the empathy generated when users kicked the robot to show it stand up again, the videos caused some public anxiety. Google concluded that Boston Dynamics isn’t likely to produce a marketable product in the next few years and put the business unit up for sale [30]. Apart of not being able to commit long term financially, Google also sought to disassociate from “scary” robots.
Acquisitions have been recurring theme in the service robot industry in part due to the immaturity of technology, and the lack of market to sustain its development. While having financial support is crucial, organizations pursuing this field should also develop intermediate technologies to sustain its progress, just as Honda. Also, by establishing a process early on, technology can be centered in achieving results and reducing the overall cost of technology. By employing systems architecture, complexity can be managed from an early stage in order to increase the chances of project survivability. Likewise, Honda could also borrow on Softbank ideas to subsidize the price of a unit, and offer Software as a Service to supplement its revenue.
The necessity of service robots is showing promise where the cost of human labor outweighs the cost of the system, and in hazardous places that put under risk human life. The 2016 Economic Report of the President (Unites States White House) [36] has predicted that many of the jobs in the \)20 per hour range have an 86 percent chance of being taken over by automation. While displacement of humans by machines have been a trend since the industrial revolution, it’s the high pace of introduction which worries officials whether humans are adapting fast enough to gain higher skill jobs. Automation has also displaced humans in a beneficial way, jobs deemed too harsh or dangerous that don’t pay enough or are not worth risk of taking but still have to be done. These could be as essential as farming, or as hazardous as nuclear facility decontamination, and even space exploration. Therefore as humanoid robots evolve and gain significance, governments should place regulations to curve deployment in order not break the delicate economic balance. Architects should also understand the downsides of their systems and be prepared to aid officials to ensure the survivability of the system.
Aligning towards success
While there is no golden template that will guarantee a successful architecture, it is worth observing the elements of past successful systems. In [12], Maeir and Rechtin highlight four lessons that successful systems have followed. Below is an interpretation on how these principles apply to ASIMO as well as well as the company.
- Right People, Right Time, Right Idea
- Honda is an engineering company which is dedicated in creating quality products. The encouragement for its employees to develop new ideas and to keep improving even the smallest detail, are some of the attributes which has led the company to building many successful systems. Their profitability has enabled the company to pour millions of dollars into the research & development of ASIMO over three decades and perhaps for many more. Projects with such a long term ambition not only need bright engineers but also long term financial support.
- The population of Japan is expected to diminish within the next decades [35]. A large percentage of the population will be in retirement which will increase the demand for senior services. Additionally, a shortage in the workforce can cause wages to increase. The demand for automation in the industrial sector continues to grow and a similar trend is expected to follow in the service area.
- ASIMO was created to assist humans with everyday tasks in households such as cleaning, cooking, gardening and caring for an elder. An extra pair of hands is always welcomed at the right price. Assuming system functionality can be met and production costs can be lowered, future demand should exist.
- Be Technically Aggressive, but not suicidal
- Most of the service robots in the market today make use of wheels as their primary form of mobility. Honda chose a biped approach because it recognized the limitation of wheels. This technology was non-existent when the program started and would require a large investment to reach an accelerated growth. Instead Honda allocated a small number of dedicate resources spread over a longer period of time to develop the proof of concept. This also allowed technological advances to catch up with the concepts that would have been nearly impossible to materialize from the beginning. Even if the program would have been rushed to completion, the system would have had a very limited capability and found itself with no market, in other words obsolete. Additionally, knowing the limitations of computing power and AI, the architecture was developed on a sensory to behavior approach with the ability to farm out the intelligence to an external system. Initially a human operator would control the system, as more autonomous capabilities are added, operator intervention can be slowly reduced. Fully intelligent autonomous systems are still not possible today; nonetheless the system is being demonstrated every day without the public noticing what goes behind the scenes.
- Consensus without Compromise
- Honda could have chosen to stay with a wheel approach just as other competitors. Removing bipedal walking would have reduced the complexity of the program, and the company could have concentrated on other areas. After all Honda is in the business of creating vehicles that operate on wheels. This decision would have clouded the overall purpose of the project and be more of a near-term solution. The environment on which the system operates would also need to be changed. However, the system was intended to be human-centric, meaning that the system should adapt around to human needs and their environment and not the other way around.
- As discussed in the previous sections, the technologies that brought ASIMO to life came from several sources. Each had to be retrofitted to take the overall architecture into consideration. For example, one challenging aspect of the system (and other major systems as well) is power. In the lab the robot can be directly hooked to an “unlimited” power source, but once operating in the real word the robot relies on its own internal source. Mechanical actions are power hungry as well as computational operations, and the current state-of-the-art batteries do not hold enough power. Combustion engines can provide more energy per mass at the cost of noise, emissions and heat, but such solution would be unacceptable for indoor operation. The size of a battery is also restricted: too small and the power is not enough, too big and the human proportion of the robot is lost.
- Architecture as Invariants
- The shape, functionality and purpose of the system have changed over three decades but the humanoid aspect still remains. From intimidating to friendly, the robot still aims to execute tasks in a similar fashion as humans would do. The system still needs to evolve to become fully functional, and by keeping focused on the same goals, the system will gradually improve.
Figure 9. Forces influencing the system
Conclusion
The role of ASIMO is to assist humans by adapting to the same environment; therefore the architecture of the system should mimic that of its representation. The important elements are mobility, manipulation, perception, communication and thought. Although the system is not designed to impart judgment, there is a certain level of intelligence needed in order to understand and execute orders from its user. Therefore an intelligent behavior model was proposed by the author because it best suits the structure of the system. Having a modular architecture not only allows some of its components/subsystems to be reused in other products, but also reintegrate back into the system with minimal risk some of the achievements found in parallel from other products. As the system is trying to break into new markets with no real customer backing the program, the architecture team has to pay attention to the public reception and mold the usefulness of the system to specific areas; all while keeping costs into consideration.
References
[0] Moon, F.C., 1939 2007, The machines of Leonardo da Vinci and Franz Reuleaux: kinematics of machines from the Renaissance to the 20th century, Springer, Dordrecht, the Netherlands
[1] Yamaguchi, Y. 2002, "All too human; Honda's walking, talking robot, Asimo, leads automaker into uncharted territory, Engineers ponder potential for sharing technology", Automotive News, vol. 76, no. 5968.
[2] Shigemi, S., Kawabe, K., Nakamura, T., 2012, “Development of New ASIMO – Realization of Autonomous Machine.” Honda R&D Technical Review, Vol.24 No.1
[3] Hirose, M. & Ogawa, K. 2007, "Honda humanoid robots development", Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, vol. 365, no. 1850, pp. 11-19.
[4] Kusuda, Y., 2002, “The humanoid robot scene in Japan”. The Industrial Robot, 29(5), 412-419.
[5] History - Robot Development Process. Retrieved from http://world.honda.com/ASIMO/history/
[6] ASIMO, Technical Information, 2007, Honda Motor Co., Ltd, Public relations Division. Retrieved from asimo.honda.com/downloads/pdf/asimo-technical-information.pdf
[7] Byko, M. 2003, "Personification: The materials science and engineering of humanoid robots", JOM, vol. 55, no. 11, pp. 14-19.
[8] Kusuda, Y. 2002, "The humanoid robot scene in Japan", Industrial Robot: An International Journal, vol. 29, no. 5, pp. 412-419.
[9] Rechtin, E., 1991, “Systems architecting: creating and building complex systems”, Prentice Hall, Englewood Cliffs, N.J.
[10] Chestnutt, J., Lau, M., Cheung, G., Kuffner, J., Hodgins, J. & Kanade, T. 2005, "Footstep Planning for the Honda ASIMO Humanoid", IEEE, pp. 629.
[11] KOSHIISHI, T., 2013, “Styling and Design of New ASIMO – Evolution for Continuing Engagement with People”, Honda R&D, Article of Honda R&D Technical Review Vol.25 No.1
[12] Maier, M., Rechtin, E., 2009, “The Art of Systems Architecting”. 3rd ed. Boca Raton, FL: CRC Taylor & Francis Group. Print.
[13] Sakagami, Y., Watanabe, R., Aoyama, C., Matsunaga, S., Higaki, N. & Fujimura, K. 2002, "The intelligent ASIMO: system overview and integration", IEEE, pp. 2478.
[14] Kim, H., Kim, J., Komatani, K., Ogata, T. & Okuno, H.G. 2009, "Target Speech Detection and Separation for Communication with Humanoid Robots in Noisy Home Environments", Advanced Robotics, vol. 23, no. 15, pp. 2093.
[15] Koike, A., Koshishi, T., 2001, “Styling Design of the Humanoid Robot ASIMO”, Honda R&D Technical Review Vol.13 No.1
[16] Normile, D., 2008, “Trumpets vs. Crumpets in a Robot Duel”. New York Times.
[17] Nikkei Report. 2009. “Interview with Masato Hirose - Falling down, getting up, and walking on". Nikkei Technology, retrieved from http://techon.nikkeibp.co.jp/english/NEWS_EN/20060623/118478/
[19] Zaun, T., 2001, “Why Did Honda Build a Humanoid Robot That Meets With the Vatican's Approval?”. The Wall Street Journal
[20] Sentis, L., Khatib, O. 2006, "A whole-body control framework for humanoids operating in human environments", IEEE, pp. 2641.
[22] Human Muscle System. Encyclopedia Britanica. Retrieved from https://www.britannica.com/science/human-muscle-system
[23] Virgala, I., Kelemen, M., Varga, M. & Kuryło, P. 2014, "Analyzing, Modeling and Simulation of Humanoid Robot Hand Motion", Procedia Engineering, vol. 96, pp. 489-499.
[24] Ferrari, F., Paladino, M.P. & Jetten, J. 2016, "Blurring Human–Machine Distinctions: Anthropomorphic Appearance in Social Robots as a Threat to Human Distinctiveness", International Journal of Social Robotics,
[25] Hesseldahl , A. 2002,”Say Hello To Asimo, Forbes, retrieved from http://www.forbes.com/2002/02/21/0221tentech.html
[26] 2005, ASIMO Has New "Home" Inside Disneyland, retrieved from http://asimo.honda.com/news/asimo-has-new-home-inside-disneyland/newsarticle_0034/
[27] Pfanner, E. 2014. “Why Softbank turned to France for an emotional robot” Japan Real Time, The Wall Street Journal.
[28] Honda Robotics. http://world.honda.com/HondaRobotics/
[30] Stone, B., Clark, J. 2016. Google Puts Boston Dynamics Up for Sale in Robotics Retreat. Bloomberg. http://www.bloomberg.com/news/articles/2016-03-17/google-is-said-to-put-boston-dynamics-robotics-unit-up-for-sale
[31] Eilperin, J., 2014. Obama finds Japanese robots ‘a little scary’. Washington Post. Retrieved from https://www.washingtonpost.com/news/post-politics/wp/2014/04/24/obama-finds-japanese-robots-a-little-scary/
[32] Shigemi, S., Kawagushi, Y., Yoshiike, T., Kawabe, K., Ogawa, N., 2006, “Development of New ASIMO”, Honda R&D Technical Review, Vol.18 No.1
[33] Hara, Y., 2002, “'Personal robots' get ready to walk on the human side”, EE Times, retrieved from http://www.eetimes.com/document.asp?doc_id=1145146
[34] Rothfeder, J., author 2014, Driving Honda: inside the world's most innovative car company, Portfolio/Penguin, New York, New York.
[35] Taylor, A., 2016. “It’s official: Japan’s population is dramatically shrinking”. The Washington Post. Retrieved from https://www.washingtonpost.com/news/worldviews/wp/2016/02/26/its-official-japans-population-is-drastically-shrinking/
[36] The Economic Report of the President 2016. Retrieved from https://www.whitehouse.gov/sites/default/files/docs/ERP_2016_Book_Complete%20JA.pdf
[1] One of the intriguing theories posited in recent years has been that Leonardo da Vinci had designed an automaton and that these designs were used to construct a walking lion and a walking knight for court entertainment [0].
[2] Mighty Atom or Astro Boy, is a child-like robot from a Japanese comic book that has super-human abilities.
[3] The human joint has one degree of freedom for each range of movement
[4] Systems architect is a relatively new term and a person may not have been referred as one before modern times
[5] Honda has not directly disclosed the price tag and its estimate of $1 million is as of 2000.
[6] Aldebaran, a French startup that created a 58 cm (23 in) tall biped humanoid robot called Nao which was popular in the research domain.
[7] Foxconn, a Taiwanese electronics contract manufacturer
[8] Alibaba, a Chinese e-commerce site that connects manufactures with consumers as well as provides payment services
[9] Boston Dynamics researches robotics legs and has engineered several four legged robots with combustion engines that can walk on a variety of terrains, sustain brute forces, and recover in case they fall.