If we can't design autonomous robots, we'll be make them design themselves.
Elon Musk's recently announced with a projected human body, "human hand" and characteristic optimistic delivery date, the with good reason. Musk says, among other possibilities, robots will eventually be able to perform errands, such as walking alone to a grocery store. Boston Dynamics, the maker of some of the most advanced humanoid robots in history, has been developing the Atlas platform for over a decade.
While Atlas has made incredible strides, including running, jumping and reaching 4,444 viewers in front of tens of millions of YouTube viewers, the company is quick to admit that robots are far from autonomously performing complex tasks. One of the best examples of breaking the promise and potential of evolutionary robotics dates back to a study published in the journal in 2010. The study authors used a physical robot equipped with motors and sensors (not just simulations) to perform several evolutionary models and fitness goals, including collision-free navigation, homing, and predator coevolution.
"These examples of experimental evolution with robots confirm the power of evolution through mutation, recombination, and natural selection. In all cases, the robots initially exhibited completely inconsistent behavior because their genomes had random values." Overall, the study concluded that "hundreds of generations of random mutations and selective breeding are sufficient to promote effective behavior in a wide range of environmental conditions."
This demand for multiple generations of evolution can be exemplified by Alphabet's recently launched more than 100 prototypes of Google's office cleaning robots. progress vs perfection I think Musk has the potential to actually beat his rival in the robotics arena, but he'll probably need the help of the robots themselves. According to many experts in the field of evolutionary computing, robots capable of performing complex tasks that require constant feedback or learning cycles are too complex for humans to build on their own. Instead, the future of robotics may be the result of an “evolution” in which robots choose the functions that are most useful to achieve specific outcomes.
Evolving robots sounds like science fiction. In the 1950s, Alan Turing hypothesized that building intelligent machines would be too difficult for human designers and that the best bet would be to introduce "mutations" and selective reproduction into the process. Of course, the idea of evolutionary robotics has been around for a long time, but the tools needed to make this concept a reality are only now available. For the first time in modern history, we have all the components needed to facilitate evolutionary robotics, including rapid prototyping and physical cloning using 3D printing, neural networks for learning and training, increased battery life and cheaper materials.
NASA, for example, has already started artificial evolution. Even more interesting is the 2020 University of Vermont and Tufts University maker "the first small biological machine created in computer simulation using evolutionary robotics techniques." This self-healing biological machine was built using stem cells from a frog and has demonstrated its ability to move and repel payloads. The idea is that these "nanorobots" could be used to deliver drugs sometime after they are injected into the bloodstream. However, despite all these findings, evolutionary iterations of physical robots are still time-consuming, in part because of the risks involved.
Even tasks like grocery shopping are incredibly difficult, and various robotic mistakes like crossing the street and entering the driveway can put people at risk. too many possibilities Musk is right when his old Tesla car is a wheeled robot, but that's an oversimplification. Tesla specializes in single tasks and the self-learning required to navigate a complex world without direct supervision is not possible. He has at his disposal supercomputers, already advanced robots, and a marvelous team of artificial intelligence experts, but he's still a long way from creating humanoid robots that can emerge on their own.
Building a robot that could operate on its own would require hundreds of "generations" of evolution in which robots mutate and combine the most desirable traits of two different parents. To dream of useful real world applications, think about security and intelligence, building safety and regulatory inspection, firefighting assistance or search and rescue assistance. In June 2021, a waterfront high-rise apartment collapsed in Surfside, Florida, killing about 100 people. A great example of the usefulness of drone swarms is building and regulatory inspection. Much more regular and frequent inspections of aging condominium buildings, from top to floor, inside and outside, can be done using sensors and cameras to ensure water resistance. Concrete problems, cracks and cracks, settlement and other problems.
This can be done at a fraction of the cost of a human engineering team. Other useful event apps include Safety & Health Care. Consider the recent Astroworld tragedy in Houston. In events with 100,000 participants, it is often difficult to cover large, densely populated areas with security personnel. In this regard, drones or swarms of robots can be very useful for tracking people tracking security issues, fights, seizures or other medical emergencies, and even bringing medical devices like external automated external defibrillator much faster than humans. ...why do drones need to be in flocks rather than just one? There are many reasons, but most of them are reliability and redundancy. If one drone fails, the operation continues without interruption.
This is especially useful in high-risk situations where the "mission" cannot be interrupted. Build the best robot The term "evolutionary robotics" is actually a bit misleading because it's about copying processes learned in organic evolution into inorganic devices. The best explanation could be "artificial evolution" or "embodied evolution". It is not a robot that evolves as much as the process itself makes it evolve. The same approach can be applied to any object to which it can be applied.