1997 Grand Challenges in Robotics

Minutes of a Panel Discussion at the International Conference on Robotics and Automation (ICRA) in Albuquerque (NM) on April 23, 1997

Organizers: Ken Goldberg (UC Berkeley)
Howard Moraff (NSF)
Panelists: Harry Asada (MIT)
George Bekey (University of Southern California)
Antonio Bicchi (University of Pisa)
John Craig (Adept)
Toshio Fukuda (University of Nagoya)
Lydia Kavraki (Rice University)
Eric Krotkov (DARPA)

US President J. F. Kennedy's promise to bring a man to the moon was a Grand Challenge in the 1960's. What are today's Grand Challenges in Robotics? Dr. Moraff summarizes properties of a Grand Challenge:

In addition to the scientific and technological issues, we have to be aware of the meta challenges which will result from achieving the goal:

Dr. Moraff identifies one particularly significant emerging area: Personal Robotics, e.g. for health care, or for care of the elderly and children. The costs saved by keeping somebody out of a nursing home justify large investments in personal robotics.

Prof. Goldberg states the need of measurable challenges and clearly identifiable goals, such as the first driverless vehicle for the highway or the first FDA approved surgical robot.

Presentations by the panelists

Prof. Harry Asada (MIT): Grand Challenges need not be as grandious as putting a man on the moon! The last ten years of human life are very important. What we need are personal robots to assist the elderly, bed-ridden, or sick. Design of a chair, bed, or toilet that can move a patient without heavy lifting by nurses. Example: design of a (water)bed with an embedded shape memory alloy actuator array able to create waves in the bed surface that move the patient. Will require distributed sensors, actuators, and human centered control.

Prof. Antonio Bicchi (University of Pisa, Italy): My (minimalistic) grand challenges are: a) To make things simpler, and b) To make things appear simpler. Simpler hardware improves cost, weight, reliability. Examples: Salisbury's WAM, 1JOC, RISC, Parallel-Jaw Manipulation by Rolling. But simpler hardware may involve more sophisticated theory! We need: advances in algorithms, analysis, control (particularly non-linear). We need also make our theoretical results "appear simpler" to industrial users, who are perhaps not theoretically-minded, but deeply aware of their application needs. Customizable benchmarks using WWW and Java appear instrumental (see e.g., FixtureNet II, The Planners' Playground , etc.).

John Craig (Adept, USA): The most important grand challenge: make robots easier to use! "two times better" does not make much difference, but "two times easier to program" does! Currently, the time from product design to production often takes 20 months. Goal: 2 months, 5 sigma reliability on the first day: 1 failure per 4000: 2000 parts per hour. This requires: CAD tools for design, optimization, and programming.;large library of components; packages for geometric reasoning; automatic program generation; good GUI.

Prof Toshio Fukuda (Nagoya, Japan): Grand challenges for pure research are okay but one big grand challenge for robotics is to make more revenue!

The new Honda Humanoid Robot which weighs 280 kg and will be revealed at IROS in September 1997.

Prof. George Bekey (USC, USA): The Grand Challenges in robotics are also human and social challenges: note the acceleration of change in society: With how much change can an average human cope? Everything SGI sells was designed within the past 5 months! Humans still age at the same rate as they did 1000 years ago. Is technological change leaving out large parts of the population? Grand Challenges:

Prof. Lydia Kavraki (Rice, USA): Grand Challenges: computational models of tolerance. This is especially important as scale diminishes.

Eric Krotkov (DARPA, USA): DARPA is having a renaissance of interest in Robotics, primarily in Mobile Robots:

Questions and Open Discussion

Shin: We need to think about interfaces for untrained users of robots.

Lumelsky (U. Wisconsin): Most of these Grand Challenges don't capture the imagination! Instead: a human-machine partnership with high intelligence in the robot. In many ways machines are smarter than humans, e.g. in logic.
Moraff: more intelligent systems lead to less undertandable systems. The challenge is to build a machine intelligence with human intuition.

Ari Requicha (USC): challenge: bottom-up assembly (from molecular level) of products. Today: with the Atomic Force Microscope we can move gold balls of 15 nm diameter. Can we manipulate DNA directly?

Minuro Asada (Osaka University): RoboCup (robotic soccer), a benchmark that promotes robotics beyond the scientific community. Important for user acceptance.
Moraff: Another challenge: amachine vision system that can beat humans in real world situations.

Yoshi Nokamura (U. Tokyo): what about human dignity and acceptance of household and service robots?
Asada: grand challenge include both grandiose problems like understanding the brain and more practical problems.

Chris Atkeson (Georgia Tech): what we need are Consumer Robots! Computation and sensing are becoming cheap, but actuation is not cheap. What about a robot that does the thinking and lets a human do the heavy lifting!
Take advantage of "world actuators" (such as gravity), distributed micro sensor systems in household appliances for early detection of heart disease, cancer ( eg: colon cancer causes iron deficiency, which can be detected by smart toilets!); smart kitchen (diet control); personalized tutoring systems.
Bekey: speaking of cancer: a Grand Challenge is a colony of micro-robots that can destroy tumor cells.
Moraff: Some Gand Challenges are much harder than they seem: machine speech recognition and language translation were considered within reach in the '70's. 25 years later: only moderate progress has been made.

Submissions via Email (ongoing!)

Christian Laugier (IMAG, France): I agree with Antonio Bicchi when he said that making things "simpler" and "easier to use" is of a prime importance: "simpler" often means more robust and cheaper, and "easier to use" is one of the first requirement for making robotics techniques acceptable by people. This is an important point, since I am convinced that one of the most deep social impacts of robotics will arise in the domain of the "every-day life" of people (I mean in applications domains like domestic tasks, services, medical, transportation ...). Most of these application domains have been mentioned and discussed during the panel session, but very few attention was given to Intelligent Tranportation Systems (ITS). I know that Robotics is only one of the components of ITS, but I am convinced that it will play a major role in future road transportation systems, provided that we will be able to produce robust, cheap enough, and easy-to-use technologies. If the proposed solutions are not acceptable for people (I mean when considering the three previous requirements on security, economy, and HMI), then we will fail.

Robin R. Murphy (Colorado School of Mines, USA): I'd like to suggest a grand challenge: build a small autonomous robot capable for Urban Search and Rescue (e.g, Oklahoma City bombing, earthquakes, etc.). Such a robot should be able to be commissioned by a fireman or other emergency response person without robotics expertise. It should be able to navigate through rubble, efficiently searching for signs of survivors and potential hazards (e.g., natural gas leaks). Such a search would almost certainly involve multiple sensors. Furthermore, the robot must be robust - ideally such a robot would literally be thrown into the rubble beyond where a human could go. It must keep operating even if it falls over (introvertible), and/or sensors are damaged. And it should return "home" before the batteries run out. I believe such a challenge would involve research and integration of work in human-centered robotics, active sensing, sensor fusion, reliability, navigation, micro-sensors, and mechanical design. Plus this is sorely needed for the "Big One" earthquake expected in LA.
Moraff: It should be fairly easy to quantify the requirements, so as to set real goals for the community to attack.

Hirohisa Hirukawa (Electrotechnical Laboratory, AIST, MITI, Japan): I'm afraid that one of major and missing building blocks to realize real useful robots is standard software library for robotic systems.

When we try to develop some robotic system, we usually must start to implement every software from scratch. To avoid such stupid situation, I agree with Antonio Bicci's idea to have standard, customizeble benchmarks on robotics software, and share it via WWW. Besides, it must be also useful to have some standard hardware or a PLATFORM on which robotics software can be developed.

From this viewpoint, MITI (Ministry of International Trade and Industry) of Japanese Government is considering to start a R&D program called 'Platform based R&D Program'. The idea is developing a hardware platform at first, distributing the platform to many research institutes, and developing standardized software on it.

A successful example of such platform based development is WinTel PC. Uncountable softwares have been developed on the WinTel platform. In robotics, we could consider the PUMA manipulator the most accepted hardware, but it is becoming obsolete. So MITI is planning to develop a humanoid type robot as a hardware platform, and enhance software development on it. This makes sense, because a humanoid has vision, manipulation and mobility and so it is possible to integrate a wide variety of software in robotics on a single hardware.

Some people may be suspicious for such a humanoid robot. But it turned out that we are able to develop a humanoid robot actually, after Honda R&D Co.Ltd. published a walking humanoid robot and shocked many people in the community. Honda will give a plenary talk on the robot at the coming IROS'97 in September. We are not sure if MITI starts the platform based R&D project or not at this point. But MITI already started two satelite projects, i.e.

These elemental technologies are expected to be applied to develop a humanoid platform in the future. Though we can't know if the platform based program really starts or not now, but it woule be great to have some cooperation with the US community or NSF when it starts, because the US community has high potential in the software.

I don't represent MITI on the program, but I'm pleased to introduce you the responsible official of MITI if you have any interest. I hope you could enjoy my proposal.

Ken Goldberg (UC Berkeley): Design and build a system that can position several hundreds of small parts (size 1mm x 1mm x 2mm) in parallel with accuracy in position of 10µm and orientation of 1mrad (0.05°). This will be useful for assembly of batch microfabricated components such as read-write heads for disk drives, or in active matrix displays.

Please send your ideas and suggestions to Karl Böhringer (karl@ieor.berkeley.edu), Ken Goldberg (goldberg@ieor.berkeley.edu), or Howard Moraff (hmoraff@nsf.gov).

Notes compiled by the Organizers with help from Karl Böhringer (UC Berkeley).
Summary of 1996 Grand Challenges Panel: http://rvsl.ece.neu.edu/pub/ras/gc.html.

Document created April 28, 1997, last modified June 1, 1997.