New Evolving Robot Teaches Itself to Walk Through Trial and Error

University of Oslo engineers have developed a new robot capable of learning how to walk by remembering and analyzing its faulty and successful steps. The robot is called Dyret for Dynamic Robot for Embodied Testing.

A new robot developed by University of Oslo engineers now has the ability to teach itself to walk by evolving through trial and error. The robot, called Dyret, is described in a study entitled "Self-Modifying Morphology Experiments with DyRET: Dynamic Robot for Embodied Testing."

"Robots need to be able to adapt to complex and dynamic environments for widespread adoption, and adapting the body might yield more flexible and robust robots," states the paper. "Previous work on dynamic robot morphology has focused on simulation, combining simple modules, or switching between locomotion modes. This paper presents an alternative approach: automatic self-reconfiguration of morphology on a four-legged hardware robot."

Dyret in action

A video provided with the paper shows the robot’s impressive skills and describes how it functions. The robot is shown adapting to new terrain.

Dyret does this by shifting its body weight and taking different steps. Often, it falls over but this too is a step in the right direction.

Motion sensors record the robot’s movements allowing Dyret to pick the successful ones. The machine is also designed to stop moving excessively when low on battery.

At that point, Dyret shortens its legs in order to move shorter limbs and preserve energy. The robot’s ability to monitor and adapt to its terrain and space means it is part of a new trend called evolutionary robotics.


This Robot Will Assist Humans with the Maintenance of other Robots

An exciting field of robotics

Evolutionary robotics is a new field of technology research, based on neo-Darwinian principles, that uses evolutionary computation to generate evolving robots capable of adapting to their environment through a process similar to natural evolution. Considering the potential applications of this field, it is a surprise that it has not evolved sooner.

According to an article in the MIT press journal, this is because evolutionary robotics has been plagued with computational issues. The authors stated the complications as: "(1) the bootstrap problem, (2) deception, and (3) the role of genomic encoding and genotype-phenotype mapping in the evolution of controllers for complex tasks."

The authors also stated that the absence of standard research practices in the field was contributing to hindering widespread adoption of the technology. They suggested potential avenues of research that could support "the establishment of evolutionary robotics as a canonical approach for the engineering of autonomous robots."

Time will tell how evolutionary robotics will develop, but for now, the field seems to at least be thriving at the University of Oslo. 

Dyret is led by Tønnes Nygaard, a roboticist with the Engineering Predictability With Embodied Cognition project at the University of Oslo. Experiments for the machine were predominantly performed in the laboratory with some preliminary tests undertaken in outdoor environments.

The Engineering Predictability With Embodied Cognition project states that its goal "is to exploit the form of various systems to develop predictive reasoning models as alternatives to traditional reactive systems."