Hans Moravec
October 2000

Freely-roaming robots that fetch, clean, guard and do other chores have been an elusive fantasy for decades. Industrial mobile robots today have a very limited market because they work only on expensively prearranged routes. Hundredfold increases in onboard computer power in the 1990s finally allowed research robots to map their own routes, and prowl research building hallways and offices daylong. Industrial applications require trouble-free months, which can probably be achieved by replacing the 2D maps in the present research machines with hundredfold-richer 3D versions - the author's main work. The likely first product within three years will be a basketball-sized camera-studded "navigation head" to be retrofitted to existing industrial transport and cleaning machines, that lets them operate autonomously in new locations simply designated. The follow-on business plan anticipates a growing industrial market to support the development of mass-market products, starting with small specialized automatic home vacuum cleaners around 2005, followed by more capable home utility robots able to manipulate objects as well as travel, and, sometime after 2010, a first generation of broadly-capable "universal robots" able to run application programs for many simple chores. These machines will have mental power and inflexible behavior analogous to small reptiles. Following decades will see an evolution through mammallike learning, primatelike imagination and humanlike abstraction. By mid-century no human task, physical or intellectual, should be beyond effective automation.

I see a strong parallel between the evolution of robot intelligence and the biological intelligence that preceded it. The largest nervous systems doubled in size about every fifteen million years since the Cambrian explosion 550 million years ago. Robot controllers double in complexity (processing power) every year or two. They are now barely at the lower range of vertebrate complexity, but should catch up with us within a half century. Here are some historical and projected robotic high points, and approximate biological analogs.

• 1950: Grey Walter tortoise Elsie
  One of eight built, with phototube eye and two vacuum tube amplifiers driving relays that controlled steering and drive motors. Exhibited very lively behavior, would dance near a lighted recharging hutch until its battery ran low, then enter. Its simple tropisms resemble bacterial "intelligence".
  
  
• 1960: Hopkins Beast
  (really 1961-63) Using dozens of transistors, the Johns Hopkins University Applied Physics Lab's "Beast" wandered white hallways, centering by sonar, until its batteries ran low. Then it would seek black wall outlets with special photocell optics, and plug itself in by feel with its special recharging arm. After feeding, it would resume patrolling. Much more complex than Elsie, its deliberate behavior can be compared to a nucleated single-cell organism like a paramecium or amoeba.
  
  
• 1970: Stanford Cart line follower
  1970: SRI Shakey blocks-world reasoner
  "Stanford Cart" line follower and SRI's "Shakey" were the first mobile robots to be controlled by computers (large mainframes doing about a quarter million calculations per second, linked to the vehicles by radio). Both used television cameras to see. The Cart could follow white lines quite reliably, Shakey could find large prismatic objects somewhat less reliably. Their control complexity was far greater than Elsie's or the Beast's (lines can be tracked using simple Elsie-like techniques with ground-mounted lights and photocells, but it takes complex adaptation and prediction to do it with ambient light from a high vantage point), and the use of computers to control robots can be compared to the advent of multicellular animals with nervous systems in the Cambrian explosion, which similarly took the lid off behavioral complexity.
  
  
• 1980: Stanford Cart 3D obstacle mapper
  A million calculation per second computer and more complex program allowed the Cart to sparsely map and negotiate obstacle courses, taking five hours to cover 30 meters, a sluglike performance.
  
  
• 1990: 2D mapping from sonar range
  Ten million computations per second and a learned sensor model permitted quite reliable 2D mapping and navigation in real time from sonar range measurements, a performance comparable to the tiniest fish, or a medium insect.
  
  
• 2000: 3D mapping from stereoscopic views       (room in motion 3D)
  2000: 3D maps of a complex scene and a person       (lab in 3D)       (person in 3D)
  A billion calculations a second and hundreds of megabytes of memory allows camera-equipped robots to build almost photorealistic dense 3D maps of their surroundings. We expect to use these techniques in commercial robots that reliably transport or clean floors or guard routes they map themselves, with guppylike intelligence.
  
  
• 2005: Projected large industrial markets for 3D-perceiving robots. A standard "head for navigation", with task-specific application software layer, can be retrofitted to existing transport, cleaning and security robots, allowing them to be taught new routes by non-specialists, and smart enough to avoid most hazards.
  
  
• 2010: Mass consumer application for 3D perception
  The first may be a small, very autonomous robot vacuum cleaner that maps a residence, plans its own routes and schedules, keeps itself charged and empties its dustbag when necessary into a larger container, working for months unattended.
  
  
• 2020: Robots develop towards universal applicability
  Larger machines with manipulator arms and the ability to perform several different tasks may follow, culminating eventually in human-scale "universal" robots that can run application programs for most simple chores. Their tens of billion calculation per second lizard-scale minds would execute application programs with reptilian inflexibility.
  
  
• 2030: Robot competence becomes comparable to larger mammals
  In the decades following the first universal robots, I expect a second generation with mammallike brainpower and cognitive ability. They will have a conditioned learning mechanism, and steer among alternative paths in their application programs on the basis of past experience, gradually adapting to their special circumstances. A third generation will think like small primates and maintain physical, cultural and psychological models of their world to mentally rehearse and optimize tasks before physically performing them. A fourth, humanlike, generation will abstract and reason from the world model.