Past 50 Yers of Robotics

Embed Size (px)

Citation preview

  • 7/28/2019 Past 50 Yers of Robotics

    1/6

    Lessons from the Past 50 Years of Robotics

    Chair: Bernard Roth

    Participants: Ruzena Bajcsy, Georges Giralt, and Takeo Kanade

    In this session the three participants presented a short overview of what they

    believe to be key issues and milestones that have brought us to the present. They

    then give their perspective as to the major issues for the future. Personally, I feel

    the major lesson from the past is that the notion of general purpose robotic devices

    proved to be too ambitious, and flawed as a generating principle. Instead, versatilespecial purpose devices have proved to be the key to successful robot

    development. I believe this is the major lesson from the past 50 years, and it will

    continue to be the case into the foreseeable future.

    Lessons Learned over 50 Years in Robotics

    By Ruzena Bajcsy

    Director CITRIS and Professor Electrical Engineering & Computer Science University

    of California, Berkeley, CA

    The way I like to evaluate what happened over time in Robotics and Intelligent

    Systems in general, is to compare how the theoretical ideas of the time matchedexperimental evidence.

    In the early 1950's the most prominent theoretical ideas relevant to Robotics

    came from control theory, that is understanding feedback, (Kalman, Wiener),

    information theory (Shannon) and behavioral psychology represented by Skinner.

    Modeling of systems was performed by analog computers. The best realization of

    these systems was based on electromechanical principles, modeling homeostatic

    behavior (Ashby). Of course, there were those who thought more digitally, as a

    paradigm for intelligent behavior and models of the brain circuitry (Turing,Rosenbluth, Pitts and MCoulagh, Minsky and McCarthy).

    In the late 60's six degree of freedom manipulators and their controllers were

    built (Scheinman, Roth, Paul). Measurements of position drove the control.

    Digital formulation of kinematical transformations of the position of each joint

    had to be invented; this in turn enabled control of these mechanisms. Other

    sensors became available, such as force sensors, which enabled control of

    P. Dario and R. Chatila (Eds.): Robotics Research, STAR 15, pp. 587592, 2005.

    Springer-Verlag Berlin Heidelberg 2005

  • 7/28/2019 Past 50 Yers of Robotics

    2/6

    dynamics. Simultaneously, mechanisms with a higher degree of freedom were

    developed and controlled (Inoue and others).

    A great revolution in Computer Vision was afforded by charge-coupled devices

    in the mid 70's. While visual information was studied independently of robotics(pattern recognition is a good example), robotics required from the beginning the

    recovery of depth information, which is naturally lost during the data acquisition

    via video cameras.

    Hence, in the early 70's structured light was invented (Will) together with a

    camera system, using triangulation principle to recover the depth information.

    Later other principles for recovery of 3D were utilized, such as stereo pairs of

    cameras, motion, monocular depth cues, and so on.

    What has changed?

    On the theoretical side, the realization that robotics needs both continuous and

    discrete models led to the new field of hybrid systems. Furthermore, the

    tremendous advances in both computational and storage capacity have supported

    new advances in developing algorithms. Perhaps even more importantly,

    completely new approaches to robotics have evolved.

    Because of the limitations of computer and storage power in the past, theapproach to sensor based control was driven by data reduction mechanisms from

    sensors. Today it is possible to use large data sets for control; hence learning data

    driven algorithms is becoming very popular. Also again for the same reasons as

    above, many more sensors distributed spatially can be considered; hence many

    more complex systems can be observed and controlled. Distributed and

    cooperative robotics is now a reality, challenging the current theoretical models

    which are by and large still more local.

    Another great advance comes from new materials and miniaturization ofsensors, processors and the means of communication, especially wireless

    communication. These systems are complex, challenging our current theoretical

    understanding. Here we have an example in which theory is lacking the

    experimentation. The recent advances from Japan in complex mechanisms, asthey are exhibited in humanoids, are an excellent example of the most complex

    systems mentioned above.

    What are the lessons?

    While in the past the theoretical ideas were ahead of experimental verification,

    today the technological advances are challenging the current theoretical

    understanding of these built systems. What I mean is that we do not have good

    predictive models of the complex systems that we built.

    588 B. Roth et al.

  • 7/28/2019 Past 50 Yers of Robotics

    3/6

    Although advances have been made in several areas, such as decision making

    under uncertainty, partial Markov models and their derivatives and game theoretic

    approaches, there is still quite a bit of unknown in the distributed systems acting in

    natural environments.The technology (both in hardware and software) changes very rapidly. There is

    not enough time or effort to study the performance of these systems, with respect

    to their robustness, reliability, maintenance, power consumption and utility under

    varied conditions. Part of the reason is that academic environments reward

    novelty at the expense of studies of the integrated system performance. On the

    other hand, the complexity of these systems is such that to develop a

    comprehensive predictive theory requires a larger group effort and stability in

    such a system. However, if these systems should penetrate our society, their

    performance will have to be guaranteed with some limits, so that the user will

    know how much to expect from such a system. This is our challenge!

    In conclusion, the most important lesson to me is to understand complex

    robotic systems. These systems are at least the same or more complex as large

    software systems. Yet, if a large software system has bugs, we may reboot thecomputer, in robotic systems, such a failure may have catastrophic consequences,

    hence the debugging and understanding of its bounds of performance are essential.

    Thoughts and Views on Robotics, the Field Status and Perspectives

    Georges Giralt

    LAAS-CNRS, Toulouse, France

    1. Robotics at the turn of the Century

    Robotics opens today, at the turn of the Century, a large perspective for seminal

    scientific and technical achievements well articulated to a highly challenging

    broad host of novel applications.

    At the theoretical level, Robotics emerges as a scientific body of concepts,methods and algorithmic tools, in fact the most challenging field in Machine

    Intelligence, which effectively interplays with a current stream of developments

    that pave the way, at the practical level, to a very large domain of novel

    applications ranging from Outer Space to Assistive and Personal Robotics.

    Industrial Robotics will pursue as a well established domain, in constant

    progress, with a large variety of market products whose development trend will

    589Lessons from the Past 50 Years of Robotics

  • 7/28/2019 Past 50 Yers of Robotics

    4/6

    rightly blend research developments within the framework of Shopfloor

    Automation.

    Non-Manufacturing Robotics concerns a wide spectrum of research-driven

    real-world cases pertaining to Field, Service, Assistive, and Personal Robotics.Machine Intelligence is here in its various themes the central research direction

    endowing the robot to act:

    (a) as a human surrogate in particular for intervention tasks in remote and/or

    hostile environments,

    (b) in close interaction with humans and operating in human environments in

    all applications encompassed by Human-Friendly Robotics, also termed

    as Human-Centered Robotics,

    (c) in tight synergy with the user, expanding into Human Augmentation.

    The subjects implied by (b) have emerged as a forefront research domain

    opening to the Grand Challenge of the Personal Robot or Robot Assistant and

    Companion.

    2. Robots and Machine Intelligence: the Intelligent Robot Paradigm

    Machine Intelligence themes encompass Advanced Sensing and

    Perception,Task Reasoning and Planning, Operational and Decisional Autonomy,functional Integration Architectures, Intelligent Human-Machine, Interfaces,

    Dependability.

    Consequently, Intelligent Robots are thus categorized in a solely computational

    capacity way as Bounded Rationality Machines, expanding on the 80's third

    generation robot definition: " (robot) . operating in the three dimensional world as

    a machine endowed with the capacity to interpret and to reason about a task and

    about its execution, by intelligently relating perception to action ".

    The above paradigm captures Machine Intelligence aspects and appears as anoperational integrated-system concept present, at various levels, in all non-

    manufacturing application domains: extreme environments and field robotics,

    public safety, unmanned vehicles and professional service robots, teleoperation

    and networking, microrobotics,., human-friendly robotics.

    Hence, we can consider a working framework which can be termed as

    Constructive Robotics where research themes are realistically instantiated andimplemented leading, as a positive step, to interesting solutions tailored to specific

    cases. Thus, it is worth to note that the integrated-system constructive approachprovides an interesting R&D setting:

    - leaves open the key basic themes and issues thus keeping right the overallresearch perspective,

    - brings in critical challenges and issues related to real world mid-termapplications and their societal and economical impact,

    - keeps a positive and reasonably effective link with industry.

    590 B. Roth et al.

  • 7/28/2019 Past 50 Yers of Robotics

    5/6

    3. The Personal Robot: a Grand Challenge

    Human-friendly robotics encompasses several front-line application domains

    where the robots operate in a human environment and in close interaction withHumans (Entertainment and Education, Public-Oriented Services, Assistive and

    Personal Robots). Right at the core of the field, emerges the forefront topic of

    Personal Robots for which three general characteristics are to be underlined: (i)

    operated by a non-professional user; (ii) that may be designed to share high level

    decision-making with the human user; (iii) that may include a link to environment

    devices and machine appendages, remote systems and operators. The shared

    decisional autonomy concept (co-autonomy) implied here, unfolds in a large set of

    cutting-edge research issues and ethical problems.

    The Personal Robot concept expanding to Robot Assistant and/or Universal

    Companion, opens a true Grand Challenge to Robotics as a scientific and technical

    field offering a mid/long term perspective to achieve a paramount societal and

    economical impact. It brings in, and questions, front-line topics encompassing

    cognitive aspects: User tunable Human-Machine Intelligent Interfaces, Perception(semantics), Learning (understanding the universe of action, user tuning),

    Decisional Autonomy, Dependability (Safety, Reliability, Communication andOperating Robustness)

    4. Mid and long-term future: expectations and looming dangers

    Some issues, perspectives and critical comments that were implied in previous

    sections:

    (i) Robot Assistant/Companion wireless links:

    - positive: operating and technical assistance, possibly distributed

    machine appendages, access to powerful remote information andexpertise sources;

    - danger: privacy issues and loss of user control on the robot,.(ii) Cognition extensions:

    - positive: compels research to address properly key issues related toperception, human-machine co-existence, learning,.

    - danger # 1: over-claiming: as usual, some of it can play a positive role

    but too much is negative and leads to damaging backlashes.

    - danger # 2: the "thinking machine" syndrome with the sequel to confuse

    wishes and "given names" with reality.(iii) Robot autonomous actions and decision-making autonomy:

    - positive: emphasizes the importance and place to implement efficient

    Machine-Intelligence in Bounded Rationality schemes and control

    processes.- danger: unwanted and, possibly, unsafe Robot actions.

    591Lessons from the Past 50 Years of Robotics

  • 7/28/2019 Past 50 Yers of Robotics

    6/6

    Dependability: the development of the capacity for a machine to act, in some

    way, independently of the control of an user, besides any other philosophical

    question, brings in the necessity to define and to implement system constraints to

    prevent unwanted, unpleasant or even dangerous behaviors (a realistic/operationalsubstitute to "Asimov's laws"!). This could be in part included as one of the main

    factors characterizing dependability for Intelligent Autonomous Robots, thus

    pointing out to the importance of this subject as a forefront research theme.

    5. Conclusion

    From Human-frontier applications such as outer-space and undersea to

    Humanoid shaped home assistants we are seeing a wide spectrum of embedded

    Machine-Intelligence Robots with currently the paramount role of two salient

    emerging vectors:

    (a) Medical Robotics as the fastest expanding field.

    (b) A Grand Challenge: the Personal Robot (Assistant, Companion)

    Paradigm.

    We shall once more emphasize the scientific aspects and the economical and

    societal impact entailed by the development of Machine-Intelligence centered

    Personal Robots. Forefront research issues range from new materials andmicro/nanotechnologies to open learning and robot dependability.

    As a concluding comment, it can be contended that we are confronted to the

    real Birth of Robotics with both the ethical and pragmatic necessity to properly

    assess the current state of the field and to clearly distinguish reality from m

    592 B. Roth et al.