Evolutionary Algorithm Spawns ‘Living Robots’ from Frog Cells

Evolutionary robotics is a computational process based on the principles of natural evolution. It deploys algorithms to search for successful robot designs in simulation. Mutations that result in designs that fit predefined criteria survive, while others are discarded.

Researchers from the University of Vermont, Tufts University, and Harvard University have extended this process by building living machines out of frog cells. The millimeter-wide creatures, called xenobots, can move, pick things up, and regenerate after harm. They are, say the team, the first living robots.

The xenobots were designed at the Vermont Advanced Computing Core (VACC), home to the DeepGreen artificial intelligence (AI) supercomputer, and Bluemoon, a multi-thousand-core, high-performance computing cluster modeled after national supercomputing centers. Said Josh Bongard, an evolutionary robotics expert at the University of Vermont, "We asked the supercomputer to design us a machine that does something very simple, which is just to walk along the bottom of a simulated Petri dish. That's it." No other criteria, such as size or shape, were defined.

The algorithm first generated a population of random robots by putting together simulated frog cells in various configurations. It then evaluated each design in a physics engine, a software program that simulates physical systems.

Many of the designs are unsuccessful, said Bongard. They don't move or they just jiggle around, "But a few might manage to move a small distance forward, so those designs survive. All the other designs are deleted by the computer."

Next, the algorithm takes the survivors and creates randomly modified copies, simulating the evolutionary process of mutation. The final result is a portfolio of successful designs— those that move the furthest and fastest in different ways.

From algorithms to life forms

At Tufts, a team led by Michael Levin, director of the Tufts Center for Regenerative and Developmental Biology, alongside microsurgery specialist Douglas Blackiston, determined which designs were most viable. The best candidates were built out of skin and heart cells harvested from frog embryos, without any genomic editing.

Then followed a process of lopping and improvement: evolving designs in the supercomputer and trying to build them. "We looped over those two phases multiple times and each time we tried to build in more from what we were learning in the biology to better inform the evolutionary algorithm," said Bongard.

One improvement was uncertainty. The biologists knew that heart cells beat together when assembled in the shape of a heart, as they are supposed to; however, the algorithm was trying to put them together in other patterns, with unknown outcomes, Said Bongard, "We built that uncertainty back into the evolutionary algorithm so that it treats these heart cells now as randomly active."

Levin said the team wanted to liberate the cells from the boundary conditions of a frog embryo. "We take them and we say to them, 'okay, now you are free to reboot multi-cellularity, you can talk to each other however you want, rearrange yourselves however you want, what are you going to make?'"

While Levin now specializes in biology, his background in computer science. He is interested in information processing in living systems, how cells compute and store memory, and the algorithms of form and function.

This plasticity, the ability of cells to come together and create new forms, connects to the hardware/software distinction in computer science, he says. "Making a frog is the default behavior of this hardware, but the software that runs on this hardware is actually much more flexible and it can do other things."

One critical consequence of the work was the emergence of new behaviors, like regeneration, "We didn't say 'make a xenobot that can recover from damage.' They just do, because the xenobots are made out of cells. Cells have 4.5 billion years of experience learning how to recover from unexpected things like damage," Bongard said.

There is no doubt that xenobots are living organisms, said Levin, as they are made of living cells cooperating to form a larger structure. "But it's a very peculiar living organism whose evolutionary history was on a computer, rather than in the biosphere."

Ultimately, Levin said, such questions (is it an organism? is it a machine?) will become meaningless. "These distinctions are going to be wiped out. I think there is going to be complete merging of all this stuff."

A new approach

This is an exciting new direction in bio-manufacturing, said Roger D. Kamm, a professor of biological and mechanical engineering at the Massachusetts Institute of Technology (MIT), who was not involved in the research. "This new approach, for the first time, leverages the emergent properties of living cell clusters and their ability to self-assemble, self-organize, and adapt to changes in their environment."

The viability of the approach has been demonstrated, said Kamm, and the resultant living systems are testbeds for further research. The same methods could be applied to more complex structures in the future. "Certainly, we have a long way to go before these find common applications, but the approach is scalable, and as methods such as bioprinting are improved and adopted for manufacturing, the approach laid out in this paper will enable rapid advances."

What might those applications be? Certainly, some will be in AI and robotics, or in medicine (for drug delivery), or in the environment. Small-scale autonomous machines acting collectively could be deployed to clean up microplastics or radioactive contamination. As Bongard points out, they are 100% biodegradable.

The xenobots' ability to heal also raises the prospect of self-renewing materials. "Our machines are terrible at this at the moment," said Bongard. "You take a sledgehammer to your laptop, and that laptop is done. It is not going to spontaneously rewire and carry on."

Unlike metal and ceramics, which Bongard described as "dumb materials", when you build machines out of living cells, you also get extra functionality, such as regeneration, at no extra charge.

Karen Emslie is a location-independent freelance journalist and essayist.