History

Evolutionary robotics has its roots in the theories of natural selection and survival of the fittest, coined by Charles Darwin. The primary objective of this field of study is to prove that robots are autonomous creatures that interact with the environment on their own accord in various methods that ultimately increase or decrease their chances of survival (Nolfi and Floreano, 2004). The prevalence of a displayed action in each subsequent generation is based on the impact of the action on the robot’s survival. The rate of survival is dependent on the robot’s ability to perform successfully during a particular task.

The field of evolutionary robotics was discussed in theory long before its establishment. Alan Turing was one of the first programmers to question the ability of machinery to evolve. Though Turing’s understanding of the potential of machines was ahead of his time, and though he believed that the theory of natural selection could apply to robots, he did not believe that humans were capable of creating robots that could evolve over time. With the initiation of the field of evolutionary robotics in the late 1990s, the doubt that Turing expressed was proven irrational.  

Programming Algorithms

Evolutionary robots are programmed with genetic algorithms and neural networks set to mimic human biological systems. Learning more about the surrounding environment causes these robots to engage in adaptive changes during their lifetime. The algorithms and networks that allow for these adaptive changes to occur are then deemed as fit and selected to be inherited by future generations (Pratihar 2003). Due to an emphasis on adaptive and observational learning, robots engage in basic behaviors without preliminary programming due to interactions with the environment and other robots.

Another type of programming used is coevolutionary programming in which robots evolve their intelligence and morphologies simultaneously by exploring relationships between other organisms and themselves. The robots themselves are altered with a more complex mechanical design, incorporation of factors that develop intelligence, and quicker life span. This will allow for a mimic of natural environments in which organisms have limited time to understand their inputs and produce appropriate outputs but have their entire life for continuous improvement (Pollack et al.,1999 ). An aspect that is heavily emphasized in these studies is the ability of the programming to be creative in nature.

A fairly new type of programming is using MGE algorithms which include transposons, which are typically present in eukaryotic organisms (Spirov, 2018). Transposons are segments of DNA that change their location in the genome to create various transcriptions. These transcriptions can promote the activation or deactivation of certain codes and thus the respective genes or the position of the transposon can cause a mutation. The goal of introducing these transposons into the programming software of the robots is to measure the effect of transposons in expanding the genome of the robots as well as understand the impact of transposons on evolution itself.

Reproduction

The process of reproduction is done artificially using human intervention. Once a specific number of robots are determined to have apt programming systems, the robots are ready to reproduce. Reproduction in this context means that the coding of the successful robots will be installed in the subsequent generation. The programming systems undergo a process similar to genetic recombination in which a random portion of the programming code is swapped between the robots. The programming is also subject to random alterations to mimic the possibilities of genetic mutation (Mitri, Floreano, and Keller,  2009). Ultimately, a filial generation will be produced in which all the robots have a similar programming code that enables all of the individuals to perform the assigned task successfully. This shows that robots could also undergo evolutionary strategies.

Types of Evolution

Robots are manipulated under various conditions and the fitness studied may belong to one of two main categories - behavior and intelligence. Even though experimenters strive to delegate the evolutionary adaptation of the robots to either cognitive or functional, it is more likely that the fitness measured uses both categories.

Functional Evolution

Observation of behavioral evolution in robots often refers to the development of self- organization within a group. Many biological organisms are equipped with sensors that can detect pheromones or other stimuli from individuals within the same population signaling the need to aggregate to achieve a common goal. Experiments testing this theory observe if robots develop a social hierarchy or structure that causes them to group together in times of low resources, demonstrating the ability of the robots to act on selective pressures (Vito, 2014). The robots are not innately programmed with features to attract or be attracted by surrounding robots; they may have sensors for vision or heat and wheels for motion. Other variables can be manipulated such as the extent to which the robots should rely upon one another. One experiment has tested this with the strategy of conditioned response in which robots are harmed if they are too close to one another but are deemed fit if they form a dynamic aggregation (Garb, 2003). Ultimately, the ability of the robots to rely on one another for survival has been observed so certain behaviors can be artificially adapted within groups of robots.

Cognitive Evolution

Robots are also designed to value the ability to outsmart competition which results in an evolutionary race to adapt intelligence. In these scenarios, the robots must not only complete the activity that is fit but it must also understand and take advantage of the weaknesses of competitors or utilize its own strengths. This correlates to the development of a brain. An first experiment required robots to collect a cube in the middle of an arena, however the robots all had different morphologies. These individual robots had to first interpret the goal, then understand how to utilize their wheels, ball joints, or other technology to obtain the cube, and finally stop other competitors from obtaining the cube. Robots with large worm like arms outsmarted the other robots by simply placing an arm over the cube, concealing it from the other robots (Floreano and Keller, 2010).

Application

At first, the application of natural selection to these robots seems rather superficial; however, these robots not only prove that Darwin’s theories are legitimate but they also allow an exploration into biological and social behaviors. By observing robots engaging in evolution, an in depth understanding of the prevalence of social species can be gained. Evolutionary robotics may also provide an answer to how intelligence is passed along generations, since there is much debate as to whether intelligence is encoded by genetic material (Silva and Christensen, 2016). The field of evolutionary robotics does not just focus on observation but also strives to implement these robots on the field. In times of environmental catastrophes, groups of robots can be released with instructions to evolve through generations to gain the most fit programming code for the environment. This would allow for robots to be prepared for the dangerous environment as well as morph when necessary to combat the changing pressure of the situation (Floreano and Keller, 2010).

​