April 30, 2001 update
The effort proposed here is underway. http://www.frc.ri.cmu.edu/~hpm, lower left corner, reports on full-time work initiated in 1999 to develop commercial prototype self-installing mobile robots by 2003.



Autonomous
Free-Ranging
Utility Robots
for Mass Market
before 2005

Hans Moravec


Carnegie Mellon University
Robotics Institute

January 1998


Overview


A brief window of opportunity is opening. For thirty years, the effort to create robots that can reliably and competently negotiate novel locales on their own has been unsuccessful. In consequence, a mass market for mobile robots other than toys has failed to materialize. Computer power and the necessary techniques to make true autonomous mobility possible are finally becoming available. Within a decade they will be widespread. The first major products of this new industry will be designed and marketed in the next several years. We propose to undertake a venture towards that end, building on our thirty years of foundational experience. An attractive form for the venture is a licensing company, to develop the technology and market it to end manufacturers.

Since 1992 we’ve developed programs to process data from robot sensors, especially stereoscopic cameras, into dense three-dimensional maps of the surroundings. These would form the core of the venture, but there remains a long list of additional issues and ideas to be explored, among them improved preliminary processing of images, automatic optimization, route planning, robot localization within maps, object classification and many applications-specific topics. These might be addressed in early in commercial work, but are also suitable for university research, as now. From the research point of view, a modest investment in this direction could have a disproportionate payback, by sparking vigorous, self-sustaining growth. A lively industry would be fertile soil for promising robotics research that today is often lost for lack of a sustaining context. There are many precedents in the computer industry. Computers, integrated circuits and the internet, once research initiatives, now grow vigorously in the open market. We judge robotics to be on the verge of an analogous flowering. Keeping the work in an open research setting as long as possible would maximize its availability to others.

The concept of using self-guiding mobile robots for transport, cleaning, and inspection has existed for most of the twentieth century, but has been realized in only modest ways. A few tens of thousands of AGVs (Automatically Guided Vehicles) are at work in factories, warehouses and other institutional locations. Many are guided by buried signal-emitting wires defining their routes, with switches signaling endpoints and collisions, a technique developed in the 1960s. More advanced models, made possible by the advent of microprocessors in the 1980s, use more flexible cues, such a optical markings on a tiled floor. The latter may use ultrasonics and infrared proximity sensors to detect and possibly negotiate their way around obstacles. The most advanced machines, manufactured in the late 1980s and 1990s, are guided by navigational markers, such as retroreflective targets at strategic locations, and by specialized site-specific programming exploiting existing features, like walls.

The newer systems can be installed with less physical effort, but all require the services of a specialist to program the initial setup, and for layout and route changes. The expense, inconvenience, inflexibility and lack of independence stemming from the elaborate setup greatly reduces the potential market for these systems. Only very stable and high-value routes are candidates, and the high cost reduces any economic advantage over human guided vehicles. Fully autonomous robots, which could be simply shown or led through their paces by nonspecialists, then trusted to execute their tasks reliably, would have a far greater market.

A generation of fully autonomous mobile robots, which navigate without route preparation, has emerged in research laboratories worldwide in the 1990s, made possible by more powerful microprocessors. The majority of these use sonar or laser rangefinders to build coarse two-dimensional maps of their surroundings, from which they locate themselves relative to their surroundings, and plan paths between destinations. The limited information in horizontal maps allows for certain ambiguities, and when navigating in new areas, these machines have a mean time between difficulties (getting lost, stuck, or even falling) measured in hours. Previous generations of commercial robots were not accepted by customers until the between failure time exceeded several months. It appears unlikely that two-dimensional autonomous navigation can meet this standard.

Our own research mobile robot research , ongoing since the 1970s, has investigated sparse three-dimensional models and dense two-dimensional maps, using camera, sonar and laser sensors. In the last five years, taking advantage of the most powerful microprocessors, we have developed efficient programs that maintain three-dimensional volumetric maps of a robot’s surroundings, containing about a thousand times the information of two-dimensional maps. Our main sensors have been inexpensive video cameras. We judge that this technique is more than adequate to be the core of a commercially viable autonomous mobile robot navigator, the first in our thirty years of work. As a bonus, the three dimensional maps produced can be used in the robot to recognize doors and furniture-sized objects in the surroundings. We have demonstrated the core techniques, but several person-years of research and development are still needed to produce even a complete laboratory prototype. This is a good time to start a focused effort, as sufficiently powerful research computers are just becoming available. With 1,000 MIPS (millions of instructions per second), which should be available in high-end personal computers in 1998, our programs can digest a glimpse of the world in less than a second, which should be adequate for slow speed indoor robots. The same 1,000 MIPS should be available in compact low cost microcontrollers, suitable for use in even small robots, before 2005.

A projected first commercial product from this effort is a basketball-sized “navigation head” to be retrofitted on existing robots, providing them with full autonomy. It would contain 360 degrees of stereoscopic cameras and other sensors, an inexpensive inertial navigation system to inform it about small motions without depending on accurate odometry from robot wheels, 1,000 MIPS of computational power, software for basic navigation, and software hooks for applications specific programming and hardware interfaces for vehicle controls. Offered as an OEM product to existing AGV manufacturers and others, it could quite possibly expand the market for AGVs tenfold from the current tens of thousands.

A possible first product with mass-market potential is a small robot vacuum cleaner, which can reliably and systematically keep clean designated rooms in a home following a simple introduction to the location.

The simple vacuum cleaner may be followed by larger and smarter utility robots with dusting arms. In subsequent products, arms may become stronger and more sensitive, to clean other surfaces. Mobile robots with dexterous arms, vision and touch sensors will be be able to do various tasks. With proper programming and minor accessories, such machines could pick up clutter, retrieve and deliver articles, take inventory, guard homes, open doors, mow lawns or play games. New applications will expand the market and spur further advancements, when existing robots fall short in acuity, precision, strength, reach, dexterity or processing power. Capability, numbers sold, engineering and manufacturing quality, and cost effectiveness should increase in a mutually reinforcing spiral. Ultimately the process could result in “universal robots” which can be do many different tasks, orchestrated by applications-specific programs.


References:

current work:
Robot Spatial Perception by Stereoscopic Vision and 3D Evidence Grids, CMU Robotics Institute Technical Report CMU-RI-TR-96-34, September 1996. also Daimler Benz Research, Berlin, Technical Report, 1996.

foundational work:
with Martin C. Martin, Robot Evidence Grids, CMU Robotics Institute Technical Report CMU-RI-TR-96-06, March 1996.

speculative design