‘Our robot is practically ‘born’ knowing nothing about its leg anatomy or how they work.’
STUTTGART, Germany — A robot dog with reflexes that teach it how to walk in an hour has been built by scientists. Researchers in Germany say the canine creation, called Morti, learns to walk quickly because it makes good use of its virtual spinal cord.
The German team built the fast-learning four-legged friend in a bid to find out more about how animals in nature learn to walk. Animals are born with muscle coordination networks in their spinal cord but learning precisely how to use their leg muscles and tendons can take time.
Baby animals begin their lives relying on hard-wired spinal cord reflexes. More basic motor control reflexes also help the animal avoid falling and hurting themselves during their first attempts at walking.
Animals must then practice more advanced and precise muscle control until the nervous system adapts to the young creature’s leg muscles and tendons.
“As engineers and roboticists, we sought the answer by building a robot that features reflexes just like an animal and learns from mistakes,” says study first author Felix Ruppert from the Max Planck Institute for Intelligent Systems in Stuttgart in a university release.
“If an animal stumbles, is that a mistake? Not if it happens once. But if it stumbles frequently, it gives us a measure of how well the robot walks.”
How does the robot dog learn this skill?
The robot dog works by using a complex algorithm that guides how it learns. Information from foot sensors is matched with data from the model spinal cord which is running as a program in the robot’s computer.
The robot dog learns to walk by constantly comparing set and expected sensor information, running reflex loops, and adapting the way it regulates its movements. The algorithm adapts control parameters of a Central Pattern Generator (CPG).
In humans and animals, these are networks of neurons in the spinal cord that produce periodic muscle contractions without input from the brain. The pattern generator networks help us walk, blink, and digest food.
Reflexes are involuntary actions triggered by hard-coded pathways that connect sensors in the leg with the spinal cord. As long as an animal walks over a perfectly flat surface, these pattern generators can be sufficient to control the movement signals from the spinal cord.
A small bump changes the walk, reflexes kick in and the creature may have to change its movement patterns in order to avoid falling. These changes are reversible and “elastic,” and movement patterns return to their original configuration after the disturbance.
If the animal does not stop stumbling, despite active reflexes, then the movement patterns must be relearned and made irreversible. In a newborn animal, these pattern generators are not yet adjusted well enough and the animal stumbles around, both on even and uneven terrain.
The creature quickly learns how the pattern generators and networks control leg muscles and tendons. The same is true of the robot but it learns its movements even faster than an animal.
Morti learns to avoid stumbling
Morti’s pattern generators are simulated on a small and lightweight computer that controls the motion of the robot’s legs. This virtual spinal cord is placed on the quadruped robot’s back where the head would be. During the hour it takes for the robot to walk smoothly, sensor data from the robot’s feet are continuously compared with the expected touch-down predicted by the robot’s pattern generator.
If the robot stumbles, the algorithm changes how far the legs swing back and forth, how fast the legs swing, and how long a leg is on the ground. As the robot learns, the pattern generator sends adapted motor signals so it stumbles less and learns how to walk.
In this framework, the virtual spinal cord doesn’t know anything about the robot’s leg design or its motors and springs. Knowing nothing about the physics of the machine means it lacks a robot “model.”
“Our robot is practically ‘born’ knowing nothing about its leg anatomy or how they work,” Ruppert explains.
“The CPG resembles a built-in automatic walking intelligence that nature provides and that we have transferred to the robot. The computer produces signals that control the legs’ motors, and the robot initially walks and stumbles,” the study author continues.
“Data flows back from the sensors to the virtual spinal cord where sensor and CPG data are compared. If the sensor data does not match the expected data, the learning algorithm changes the walking behavior until the robot walks well, and without stumbling. Changing the CPG output while keeping reflexes active and monitoring the robot stumbling is a core part of the learning process.”
‘The robotic model gives us answers to questions that biology alone can’t answer’
“We can’t easily research the spinal cord of a living animal. But we can model one in the robot,” adds study co-author Alexander Badri-Spröwitz.
“We know that these CPGs exist in many animals. We know that reflexes are embedded; but how can we combine both so that animals learn movements with reflexes and CPGs? This is fundamental research at the intersection between robotics and biology. The robotic model gives us answers to questions that biology alone can’t answer.”
The authors say their robot is good for the planet as it uses only five watts of power to walk.
The findings are published in the journal Nature Machine Intelligence.
South West News Service writer Gwyn Wright contributed to this report.