Skip to Main Content
Skip Nav Destination

Evolving robots could optimize chemotherapy

18 March 2022

The puck-shaped autonomous devices can breed, die, and be reborn as they roll about a table in search of life-sustaining resources.

3 cylindrical robots with LEDs on top on a green background
Credit: Gao Wang

Chemotherapy disrupts cancer cells’ ability to reproduce by frustrating cell division and damaging the cells’ DNA. In response to the pharmaceutical onslaught, cancer cells acquire mutations that reduce the therapy’s effectiveness. Compounding the challenge of fighting cancer: Under chemical and other stresses, mutation rates increase.

A team led by Princeton University’s Robert Austin and Chongqing University’s Liyu Liu has developed a novel approach to study—and potentially thwart—cancer cells’ adaptation to chemotherapy. Their cancer cell analogues are wheeled, cylindrical robots about 65 mm in diameter and 60 mm in height (see photo above). Fifty of the robots roll independently of each other over a square table, whose 4.2 × 4.2 m2 surface is covered by 2.7 million LEDs (see photo below). Light from the LEDs serves as the robots’ food. Once a robot has “eaten” the light beneath it, the corresponding LEDs are dimmed until they recover a fixed time later.

The bottom surface of each robot is equipped with four semiconductor-based sensors that can detect the intensities and spatial gradients of the three colors of light emitted by the light table: red, green, and blue (RGB). Each robot’s six-byte genome analogue determines how sensitive it is to the three colors. The sensitivity, in turn, determines how readily the robot moves in response to the colors’ intensities and spatial gradients.

Like humans, the robots inherit genes from two parents (how robots exchange genes is discussed below). And, as is the case for some human genes, a robot’s sensitivity to each RGB color can be dominant or recessive. If it’s dominant, the sensitivity can take on an integer value from 1 to 127. If it’s recessive, the sensitivity is zero—that is, the robot is completely insensitive to that color.

Using uppercase to denote a dominant gene and lowercase to denote a recessive gene and using pairs of letters to denote genes from the two parents, a particular robot could have the genotype Rr gg bb. That robot would be sensitive only to red light, and it could consume only red light as food. A robot with the genotype Rr Gg Bb would be sensitive to and consume light of all three colors. A robot with the genotype rr gg bb would be unable to move in search of food. It would starve and die.

Besides the RGB sensors, each robot also has eight IR LEDs and four IR sensors. The IR components help the robots recognize each other and effect gene transfer. RGB LEDs on the top of each robot display genetic information. If a robot has the genotype Rr gg bb, only the red LED glows. If it has Rr Gg Bb, all three LEDs glow. A robot with rr gg bb does not glow at all.

Light from the robots’ RGB LEDs is tracked by a wide-angle camera that looks down on the robot colony and records its state, including the positions of each individual robot. Information collected by the camera is also used to dim LEDs once a robot has fed on them. That information collected by the camera constitutes the observational data that Austin and Liu’s team members analyze and interpret.

Death is not terminal in the robot colony. If a starved dead robot is visited in turn by two different alive robots, it acquires genes from each one via the exchange of IR signals and returns to life. The revived robot’s new genotype contains a mix of genetic material from its two parents. For example, a revived robot could have parents that have the genotypes Rr Gg bb and Rr gg bb. Using Roman font to denote material from the first parent, the offspring could end up with the genotype rr Gg bb.

As the robots graze on the light table, their genes mutate at a rate that’s determined by the stress of finding food. An omnivorous robot with the genotype Rr Gg Bb that found itself on top of a patch of bright white light (all three colors) would face little stress. Its mutation rate would be low. Conversely, a robot with Rr gg bb that eats only red light would face high stress on top of a patch of blue light. Its mutation rate would be high.

A table of dozens of light-emitting cylindrical robots
Credit: Gao Wang

One challenge the team faced was adjusting the system’s various parameters to yield results that are both interesting and relevant. For example, if the robots exchange genes too slowly or if the depleted light takes too long to recover, the robot colony will become extinct before its collective behavior reveals any insights. To avoid that outcome, the team developed a theoretical model. Besides guiding the selection of parameters, the model identified two extreme cases where survival is high: when the robots mutate fast enough to avoid death and when they barely mutate, and death and rebirth balance. Between the two cases is a survival minimum—dubbed the meltdown valley—which leads quickly to the colony’s total extinction.

Austin and Liu’s team can set the light table to display an arbitrary mix of colors that changes spatially and temporarily. But even a steady, uniform environment of white light elicits rich behavior. That’s because the robots’ feeding changes the environment in a way that depends on their genotypes and their mutation rates.

In their paper, the team reported the results of three experimental runs of varying environmental complexity. Robots in the first experiment inhabited a fixed white environment. As was the case with the model, the colony’s survival probability featured a meltdown valley. The second experiment featured an environment whose colors changed periodically with time but remained spatially uniform. The robots adapted to the landscape, but the meltdown valley widened and deepened.

The environment in the third experiment changed randomly in both space and time. The robots struggled to adapt. Analyzing the robots’ genotypes revealed why. Despite the high mutation rate induced by the shifting environment, the robots’ genetic diversity plummeted. The robots became less adaptable. That finding was not predicted by the theoretical model.

Another surprise came from the RGB sensors. A robot that could ostensibly detect light of only one color could also detect light from a spectrally neighboring color. The robots, it turned out, responded to resources that they neither consumed nor needed. In biology, the ability of a gene to influence two or more unrelated physical traits is known as pleiotropy. It has been implicated in cancer cells’ resistance to chemotherapy.

The robots’ evolution suggests a way to improve chemotherapy. Instead of periodic doses of a single drug, stochastic doses of multiple drugs could be more effective. (G. Wang et al., Proc. Natl. Acad. Sci. USA 119, e2120019119, 2022.)

Close Modal

or Create an Account

Close Modal
Close Modal