A Fully Autonomous Robot Builds Its Own Brain and Learns from Scratch

For the first time in history, a robot has built its own synthetic central nervous system and then learned not only to walk, but how to autonomously enter and navigate the corridors of complex buildings.

  • Share on TwitterShare on FacebookShare on Google+Share on LinkedInEmail a friend

(PRWEB) September 19, 2004

This dramatic experiment was recently conducted at Imagination Engines, Inc. (IEI) in St. Louis, Missouri. Company President & CEO, Dr. Stephen Thaler points out that heretofore, scientists in the field of artificial intelligence have grossly over exaggerated claims that their robots are autonomous when in fact, immense scholarly efforts have been poured into writing what he calls “if-then-else” computer programs. Alternately, he points out that genetic programmers have devised schemes wherein neural circuitry evolves to enable robots to perform moderately challenging tasks. However, close analysis of the engineering results reveal that these feats are not so amazing, nor are they accomplished in convenient time scales. Tasks as simple as navigating a simple racetrack maze typically requires about 48 hours, not to mention the month invested in writing and perfecting the underlying computer program!

In stark contrast, Thaler and his assistants simply sit back, fold their arms and watch neural networks spontaneously connect themselves in a matter of seconds into the neural circuitry required for extremely ambitious robotic brains. The resulting neural network architecture both resembles and functions like a brain, a collection of individual neural networks fused into a contemplative system that can form complete models of their worlds, consider alternative scenarios, and finally choose that alternative best suited to a given problem. …Thaler quickly points out that the neural circuitry developed through genetic programming are only “reactive.” They are tantamount to reflex reactions in the brain or spinal chord wherein a stimulus simply triggers a response. The self-forming brains of IEI’s robots are entirely different. Like human brains, they think, experiment, and automatically perfect their behaviors to produce downright unexpected results, what can only be called creativity.

Even more exciting is the methodology used to accelerate the learning and the bootstrap toward creative behaviors. Rather than carry robots to various world settings and wait for different kinds of challenging scenarios to arise in order to enrich their learning experience, IEI’s robots effectively go to sleep and enter a virtual dream world wherein they experience myriad settings and situations against which they may pit their accumulating knowledge and creativity. When they awake, they may apply their dream state training to the real world, or use such experience to devise even more ambitious responses to newly arising environmental scenarios.

Recently, in a dramatic experiment conducted for the DoD, IEI scientists and engineers built a complex hexapod robot that effectively began life as a kind of “cybernetic road kill,” essentially a heap of tangled legs and electronics that learned how to walk in a period of only minutes. Continuing its learning in virtual reality, it self-originated new methodologies for navigating complex facilities and landscapes, as well as novel kinds of locomotion wherein it assumed bipedal stances to quickly evade threats. Awakened from its virtual reality test environment, it could then carry out similar behaviors in reality. The military is likewise considering such creative robots as sensor platforms for force protection and urban warfare scenarios. Visionary military thinkers see them fulfilling roles ranging from that of brilliant swarm munitions to the fully autonomous neural network based cyber-warriors anticipated by science fiction.

All of this truly revolutionary robotic technology is based upon IEI’s expansive suite of fundamental neural network patents.

# # #


Contact