Robot dog with ‘virtual spinal cord’ learns to walk in just one hour

Scientists have built a four-legged, dog-sized robot with a “virtual spinal cord” that has learned to walk from scratch in just one hour, an advance that also sheds light on the biology behind locomotion in newborn animals.

The quadruped robot Morti, described in the journal Nature Machine Intelligence on Monday, optimises its movement patterns faster than an animal, learning to walk in about one hour, scientists say.

In animals, muscle coordination networks located in the spinal cord help them make the first steps, but learning the precise coordination of leg muscles and tendons takes some time, researchers from the Max Planck Institute for Intelligent Systems (MPI-IS) in Germany, say.

Baby animals initially rely heavily on hard-wired spinal cord reflexes, studies have shown.

The labrador-sized robot has reflexes just like an animal and learns to walk from its mistakes using its complex leg mechanics and a learning algorithm, says Felix Ruppert, a formal doctoral student at MPI-IS.

“We can’t easily research the spinal cord of a living animal. But we can model one in the robot,” study co-author Alexander Badri-Sprowitz said in a statement.

In the quadruped robot, data from the foot sensors are constantly matched with target data from its modeled virtual spinal cord running as a program in its computer.

It learns to walk by continuously comparing the sent and expected sensor information, running reflex loops, and adapting its motor control patterns, researchers explain in the study.

The robot has a Central Pattern Generator (CPG) that works similar to networks of neurons in animal spinal cords that produce periodic muscle contractions without input from the brain.

These nerve networks aid the generation of rhythmic tasks such as walking, blinking, or digestion in animals.

When young animals walk over a perfectly flat surface, they say these networks of nerves can be sufficient to control the movement signals from the spinal cord, but a small bump on the ground can change the walk.

This is when reflexes kick in and adjust the movement patterns to prevent the animal from falling.

But in newborn animals, researchers say these nerves are not adjusted well initially enough and the animals stumble around, but they soon learn how these reflexes control leg muscles and tendons.

“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?” Dr Badri-Sprowitz said.

Scientists say Morti’s CPG – simulated on a small, lightweight computer, controlling the motion of the robot’s legs – also learns the same way.

Sensor data from the robot’s feet are continuously compared with the expected touch-down predicted by this virtual spinal cord.

“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,” Mr Ruppert explained.

“Changing the CPG output while keeping reflexes active and monitoring the robot stumbling is a core part of the learning process,” he added.

The learning algorithm changes how far the legs swing back and forth, how fast the legs swing, and how long a leg is on the ground when the robot dog stumbles, the study noted.

“Our robot is practically ‘born’ knowing nothing about its leg anatomy or how they work. The CPG resembles a built-in automatic walking intelligence that nature provides and that we have transferred to the robot,” Mr Ruppert added.

“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,” Dr Badri-Sprowitz added.