A Blog by Jonathan Low


Jan 9, 2022

The Reason Robots Will Have To Design the Next Generation of Themselves

The anticipatory and competent robots that humans have been led to believe will emerge by the Jetsons and Star Wars are now believed by robotics experts to be too complex for humans to design. 

Robots themselves will have to self-evolve in order to identify the traits most useful to improving their own performance. Should be fun to watch...JL

Gideon Kimbrell reports in Tech Crunch:

Robots capable of complicated tasks that require constant feedback are too complex for humans to design on their own. The future of robotic development could be “evolution” that has robots selecting which features are most useful for a specific outcome. A few hundreds of generations of random mutations and selective reproduction were sufficient to promote the evolution of efficient behaviors in a wide range of environmental conditions. Experimental evolution with robots verify the power of mutation, recombination and natural selection. In all cases, robots initially exhibited uncoordinated behavior because their genomes had random values.

Elon Musk’s recent announcement of an upcoming Tesla Bot — complete with a human form, “human-level hands” and a characteristically optimistic delivery date — has garnered a healthy serving of criticism for good reason.

Among other capabilities, Musk says, the robot will eventually be capable of running errands such as going to the grocery store alone. Boston Dynamics, which has developed the most advanced humanoid robot ever created, has spent more than a decade working on its Atlas platform. While progress has been impressive, with Atlas running, jumping and even dancing in front of tens of millions of YouTube viewers, the company is quick to acknowledge that the robot is a long way from performing complex tasks autonomously.

One of the best examples of evolutionary robotics potential — and unfulfilled promise — goes as far back as 2010 to a study published in the PLOS Biology journal. The study’s authors used physical robots equipped with motors and sensors (not just simulations) to conduct several evolutionary models and fitness goals: collision-free navigation, homing, predator-prey coevolution and more.


They concluded that “these examples of experimental evolution with robots verify the power of evolution by mutation, recombination and natural selection. In all cases, robots initially exhibited completely uncoordinated behaviour because their genomes had random values.”

In sum, the study concluded that “a few hundreds of generations of random mutations and selective reproduction were sufficient to promote the evolution of efficient behaviours in a wide range of environmental conditions.”

That requirement of so many generations of evolution is illustrated by Alphabet’s recent release of more than 100 Everyday Robot prototypes to perform cleaning chores around the Google offices — with their awkward and halting movements, the machines are still very much a work in progress.

Progress versus perfection

I think there’s a chance Musk could actually leapfrog the competition in the field of robotics, but he’ll need some help from the robots themselves. According to many experts in the evolutionary computation space, robots capable of complicated tasks that require constant feedback or learning loops are simply too complex for humans to design directly on their own. Instead, the future of robotic development and design could be a product of “evolution” that has the robots selecting which features are most useful for a specific outcome.

Evolutionary robotics sounds like sci-fi, but it isn’t a new concept. Even as early as the 1950s, Alan Turing postulated that the creation of intelligent machines would be too complex for human designers and that a better method might be introducing “mutations” and selective reproduction into the process. Of course, while the idea behind evolutionary robotics was taking shape long ago, the tools necessary to put the concept into action have only now become available.

For the first time in modern history, we have all the necessary building blocks to facilitate evolutionary robotics: rapid prototyping and physical reproduction using 3D printing, neural networks for learning and training, improved battery life and cheaper materials and much more.

NASA has already deployed artificial evolution to develop antennas for satellites, for example. Even more exciting than that, creators at the University of Vermont and Tufts University in 2020 unveiled “xenobots,” which are “tiny biological machines first designed in computer simulations using the techniques of evolutionary robotics.”

These self-healing biological machines were built using frog stem cells, and they exhibited the ability to move and push payloads; the thought is that these “nanorobots” could one day be used to deliver drugs after being injected into the bloodstream.

But even with all these breakthroughs, evolutionary iterations in physical robots remain time-consuming, partly because of the risk involved. Even a task like going grocery shopping is deceivingly complex, and a variety of robotic mistakes like crossing a street in front of traffic could put humans in danger.

So many possibilities

Musk is correct that his existing Tesla cars are simply robots on wheels, but it’s a gross oversimplification. Teslas are specialized for a single task and incapable of the self-learning necessary to navigate a complex world without direct supervision. He may have at his disposal a supercomputer, already advanced robots and a phenomenal team of AI experts, but delivering a humanoid robot capable of independently venturing out into public is likely a long way off.

Creating a robot that can operate on its own would likely require several hundred “generations” of evolution in which robots perform mutations and combine the most desirable traits from two different parents.

To daydream some useful real-world applications, think along the lines of security and recon, building safety inspections and code compliance, firefighting assistance or even search and rescue assistance.

In June 2021, a beachfront tower of condos collapsed in Surfside, Florida, claiming close to 100 lives. A great example of the usefulness of drone swarms would be building and code inspections: They could perform much more regular and frequent inspections of aging condo buildings — from the top floor to the bottom, inside and out — using sensors and cameras to check for waterproofing issues, concrete spalling and cracks, sinking and other problems. This could be done at the fraction of the cost of a team of human engineers.

Other useful applications for events include security and medical assistance. Think of the recent Astroworld tragedy in Houston. At a 100,000-person event, it can often be difficult to cover expansive and crowded terrain with human security personnel. A drone or robot swarm can be very helpful in this regard, monitoring for security issues, fights, people having seizures or other medical emergencies and even bringing medical devices such as an automatic external defibrillator much faster than human staff could.

Why a drone swarm and not a single drone? Quite a few reasons, but chiefly resilience and redundancy. If one drone fails, the operation continues uninterrupted. This is particularly helpful for high-risk situations in which the “mission” cannot be aborted.

Creating better robots

The term “evolutionary robotics” is a bit misleading because it’s really about replicating processes learned from organic evolution to non-organic devices. A better descriptor might be “artificial evolution” or “embodied evolution.” It’s not so much the robots that are evolving, but rather the processes themselves that are creating an evolution.

The same approach could be applied to any entity that can be equipped with a neural network and evolutionary algorithms to create “offspring” through both mutation and subsequent recombination from two or more parents. In fact, evolution doesn’t even need a physical form — these same processes can be deployed inside supercomputers to solve major problems. What could a better understanding of evolution help us accomplish?

Autonomous real-world interactions, for one. Evolutionary robotics is the only way to create robots capable of complex, autonomous real-world interaction. The benefits of such robots are too long to list, but use cases could range from robotic firefighters and search-and-rescue robots to nuclear waste cleanup robots, home care robots and more.

We could also gain a better understanding of organic evolution. A more nuanced knowledge of evolution could have such broad applications that it’s difficult to fathom. We could gain incredible insights into the best ways to treat diseases and build immunities, improve our life spans, lessen our impact on the ecological world and otherwise gain a better grasp of our future on this planet.

We could also garner clues into life’s origins. By studying and mastering artificial evolution, we’ll be able to better understand all the possible ways life could form and evolve on other planets. Although the possibility of life existing elsewhere in the universe remains low according to many scientific experts, a better understanding of evolution and the ability to replicate macroevolution on a micro scale will undoubtedly help guide us in any search for extraterrestrial life.

A double-edged sword

Finally, think about a deeper exploration of our solar system. With fully autonomous, self-replicating and evolving robots, we could send unmanned missions deep into space — farther than we’ve ever imagined. These robots would be able to adapt to whatever planet they landed on, reusing components, evolving according to their environment and eventually sending data or offspring back to Earth.

If the idea of robots roaming the streets conjures images of a “Terminator”-like robot uprising, you can take solace in the fact that a robot capable of learning, reproducing, observing its environment and evolving is still a long way from reality.

Instead, the biggest drawback to mastering truly autonomous robots capable of complex real-world interaction is the inevitable displacement of the human workforce. Musk believes the solution for this is universal basic income and that work in the future will be entirely optional.

I’m not sure I agree. Humans derive a sense of self-worth and value from working and creating, and to take that away could have far-reaching psychological impacts in addition to the potential financial fallout. It’s a complex problem, but evolutionary robotics could be one of the greatest achievements and biggest challenges humanity will have to face.


Post a Comment