Morti, the first robot dog that learned to walk on its own within an hour!

Spinal cord studies

This robot dog, called Morti by the researchers, was designed to study an important structure in the central nervous system of mammals, the spinal cord. More commonly called the spinal cord, it is inserted within the vertebral column itself in the vertebral canal. It begins in the medulla oblongata located at the level of the brainstem at the base of the brain and ends in the conus medullaris located at the level of the last lumbar vertebrae.

Like the brain, the spinal cord is made up of white matter and gray matter. The white matter located at the periphery of the brain includes extensions of neurons called axons with their myelin sheath. The gray matter, inside the brain, is made up of neurons.

The spinal cord is used to transmit nerve impulses. It carries sensory information to the brain and motor impulses to different parts of the body such as the legs for example. The spinal cord is also the coordination center for motor automatisms where it is involved in reflex reactions.

At birth, most animals learn to walk quickly. It is a question of survival because walking and especially running allows them to escape their predators. To do this, they are equipped from the moment they come into the world with a complex network of muscular coordination in their spinal cord.

At first, their first steps and movements are somewhat uncertain and clumsy and they often stumble. With a little practice, they will quickly end up coordinating their muscles and nerves to move more smoothly and faster.

(Also read: Kitchen robots)

Reflex movements without brain intervention


This video shows the robot dog Morti learning to walk. Source: Science X-Tech Xplore/YouTube

These indeterminate movements at first, followed by muscular coordination, are the result of a mechanism that exists in animals and humans. This mechanism involves a system known as the central pattern generator or CPG. It is a circuit of neurons that produces periodic muscle contractions without the intervention of the brain. It can act completely autonomously to create a movement like walking, for example. This is also the mechanism that occurs during the blinking of the eyes or the peristaltic movement of the intestine during digestion.

Shortly after birth, when the young animal tries to take its first steps, the CPG system is not yet fully functional and requires adjustments. This is why if it encounters an obstacle such as a gap during its initiation, the young animal stumbles and often falls. Fortunately, in animals and humans, learning allows for the development of walking movements.

Since this phenomenon is not widely known, scientists decided to use robotics. Engineers and roboticists have created a robot with reflexes similar to a young animal learning to walk. They equipped this robot dog with a virtual spinal cord managed by a learning algorithm.

(Also read: Me, robot and almost human: 6 challenges to overcome to imitate human)

The virtual spinal cord of the dog Morti robot powered by a learning algorithm

This learning algorithm, which in a way imitates a CPG system, allows Morti to learn to walk in just one hour thanks to the sensors attached to his feet. The information provided by these sensors is continuously compared with the basic sensory data contained in the robot’s memory.

Thanks to this algorithm that constantly monitors his paws, Morti adjusts gradually and gradually improves his motor skills with lightning speed. At the beginning of the apprenticeship, every time Morti fell, the program changed the amplitude and the speed of the swing of the legs, but also the length of time that each leg remained on the ground.

The computer that controls the walking of the robot dog Morti is energy efficient as it uses only 5 watts, which is very little compared to robots used in industry. Usually the algorithms and controllers of these machines are adapted to the mass of the robot, but also to its geometry.

Thanks to the Morti robot dog, scientists now have a model that enables them to study the function of the spinal cord, a structure of the nervous system that remains difficult to study in a living animal.

(Also read: How is this robot cheetah so fast?)

Source: Ruppert, F., Badri-Spröwitz, A., “Learning plastic matching in robot dynamics with closed-loop central pattern generators”, Nature Machine Intelligence, 4, 652–660 (2022), https://doi.org/10.1038/s42256-022-00505-4

Leave a Comment