The Robo-Dog learns to walk independently

They stumble precariously: newborn animals are clumsy at first. The robot dog now simulates how he learns to use his legs skillfully: “Morti” can optimize his movement patterns using an adaptive computer system that acts like an artificial nervous system. In just one hour, the robot learns to move smoothly by itself. According to the researchers, the concept can therefore be used for basic research at the interface between robotics and biology.

This is especially true for foals: newborn animals must first learn to use their muscles and tendons in a coordinated manner in order to be able to move forward confidently. In early attempts at walking, initially only reflex-based stabilization movements protect young animals from sudden falls. On the other hand, better muscle control must be developed first. Basically the following emerges: The so-called central generator of movement patterns in the spinal cord is trained during bumpy walking attempts until motion control finally allows the young animal good control of the legs. But exactly how this happens is unclear.

“We cannot examine the spinal cord of a living animal very well. But it can be modeled in a robot, ”says Alexander Badri-Spröwitz from the Max Planck Institute for Intelligent Systems in Stuttgart. “We basically know that animals have a central generator of movement patterns and reflexes. But how are they combined in such a way that they can learn by means of reflexes and a pattern generator? ” To develop a system to study this question, Badri-Spröwitz and his colleague Felix Ruppert built the four-legged robot “Morti”. “It’s a system that has animal-like reflexes and learns from mistakes,” says Ruppert.

The learning algorithm trains the virtual spinal cord

Morti is equipped with a movement pattern generator in the form of a small computer that controls the artificial muscles and tendons in the legs and thus corresponds to his biological model. In humans and animals, the network of nerve cells in the spinal cord forms this system that causes the muscles to contract rhythmically without being influenced by the brain. As long as the young animal is moving without any problems, this controller sends its basic traffic signals unchanged. This changes when the animal stumbles. Then reflexes come in and adjust the motion pattern so it doesn’t fall off. The researchers explain that if an animal repeatedly stumbles in processes that are essentially suboptimal, the learning processes can lead to a permanent adaptation of movement patterns.

Now they were able to give Morti exactly this skill. It was possible thanks to the so-called Bayesian optimization learning algorithm, which influences the development of a traffic pattern generator, which was initially not optimally adjusted. The information from the foot sensors is compared with target data from the computer – the virtual spinal cord. The researchers explain that the robot is learning to walk better and better by constantly adjusting the structure of the movement patterns sent to the sensor information. Specifically, this means that if the robot stumbles, the learning algorithm changes the forward and backward movement of the legs, the speed at which they move, and the length of time the legs remain on the ground. The “updated” motion pattern generator then sends out signals that allow the robot to walk with as little stumbling as possible: Morti gradually learns to walk more skilfully.

A sure walk after an hour

As it turned out, the robot is even slightly better than its biological role models: unlike young animals, Morti develops a safe locomotion pattern from the initial bumpy treadmill walk within an hour. The optimization is also reflected in the energy of the four abdomens when running, the researchers say: By making better use of the advantages of its mechanics, the robot improves its energy efficiency by 42 percent. Ruppert and Badri-Spröwitz have succeeded in developing a model for learning to walk by animals. “Our robot was born as a symbol and knows nothing about how its legs work. Our system works like the built-in automatic walking intelligence that nature provides and that we have transferred to the robot, ”concludes Ruppert.

As both scientists explain, the system can now be further developed and thus used for basic research at the interface between robotics and biology. Because their system is of course also interesting from a technical point of view: the computer in Morti only consumes five watts. Scientists say that other adaptive systems with comparable performance require significantly more energy. But most of all, Morti introduces nature: “The robot model can give us answers to questions that biology alone cannot answer,” says Badri-Spröwitz.

Source: Max Planck Institute for Intelligent Systems, article: Nature Machine Intelligence, doi: 10.1038 / s42256-022-00505-4

Leave a Comment