CPGs are networks of spinal neurons that generate the rhythmic patterns of neural activity which control locomotion. I wrote about them earlier this year, in the context of the “robo-salamander” designed and built by Auke Jan Ijspeert and his colleagues, of the Biologically Inspired Robotics Group at the Ecole Polytechnique Federale de Lausanne in Switzerland.
The robo-salamander is a modular robot, as it consists of identical units linked together in a chain. The adaptive furniture project is based on this technology, and, in fact, Ijspeert is one of the project leaders.
In her article, Bains covers the robo-salamander work by Ijspeert’s group, and goes on to discuss how robotic locomtion based on CPGs is being used to develop the “Roombots”. Below is my article on robo-salamander, which I posted on my old blog back in March, when Ijspeert’s group published their work in Science.
The phrase “running around like a headless chicken” is based on the observation that a chicken can flap its wings or run around frantically for several seconds after being decapitated. These movements are executed by the brain, and controlled by neural circuitry in the spinal cord, where networks of neurons, called central pattern generators (CPGs), generate rhythmic patterns of neural activity. The patterns can persist autonomously, with very little direct input from the brain, and can therefore continue when the head is severed.
It is by mimicking CPGs which form the locomotor circuit in the spinal cord of the salamander, Auke Jan Ijspeert and his colleagues, of the Biologically Inspired Robotics Group at the Ecole Polytechnique Federale de Lausanne in Switzerland, together with collaborators at the University of Bordeaux, in France, designed and built Salamander robotica, a neuromechanical model of the salamander. The robo-salamander is an amphibious robot that can crawl, walk and swim. The movements are generated by an algorithm which simulates the patterns of activity generated by the CPGs in the salamander spinal cord. Instructions are sent wirelessly from a laptop to the robot, which consists of 9 rigid, connected links, that represent the salamander’s neck, trunk and tail, and four other links, each representing a limb.
The salamander was chosen as the experimental model because its central nervous system is very similar to that of the lamprey, a primitive fish which has been studied extensively. The neural circuitry controlling swimming in the lamprey is well understood, and therefore served as a basis for the computational model used in developing the robot. Salamanders are also believed to be similar to the first organisms that emerged from the oceans to walk on land. This event, a crucial one in the evolution of the vertebrate lineage, is believed to have occurred during the Devonian period (408-360 million years ago) of the Palaeozoic era.
Being an amphibian, the salamander can both swim in water and walk on land. Both types of gait are generated nerve networks (the CPGs) located entirely within the spinal cord. CPG neurons generate both types of gait by producing co-ordinated patterns of rhythmic neural activity and sending it to the muscles.
The salamander’s swimming gait is very similar to that of the lamprey: the limbs are folded backwards against the side of the body, and undulations of the entire body propel it forward. The movements are generated by travelling waves of neural activity, which begin at CPGs near the head, and are propagated along the length of the body to the tail. The wavelength is the same as the length of the body, and nerve cells in the CPG fire sequentially, activating the circumferential muscles they innervate as the wave travels along the body. As the muscles on one side contract, and those on the other relax, that region of the body bends to one side. The wave of neural activity can be initiated experimentally by placing isolated salamander and lamprey spinal cords in a solution containing NMDA (which suggests that ionotropic glutamate receptors are involved).
For walking, the pattern of neural activity is somewhat more complex, because the body bends in two places and in opposite directions, and diagonally opposed limbs are in phase with each other (the left forelimb is in phase with the right hindlimb, and vice versa). The limb movements cause S-shaped waves of the body, which bends from one side to the other during walking, and co-ordination of the limb and body movements increases the length of the strides. This gate is generated by oscillations of synchronized activity in different parts of the CPG.
Previous work by the same group showed that a salamander’s gait could be switched from walking to swimming by increasing the amount of electrical stimulation applied to a specific region of the brain stem. The more this area of the brain stem was stimulated, the faster the salamander wiggled its body from side to side. At a specific point, there was a seamless transition from the walking to the swimming gait. So it seemed that the major difference between the swimming and the walking gait was the frequency of the body movements.
This led Ijspeert and his colleagues to speculate that the salamander’s gait is controlled by two CPGs, each one controlling gait-generating movements of different frequencies. Thus, one of the nerve networks generates low frequency body movements which result in walking. But once walking movements reached a specific frequency, the CPG responsible for generating the gait is switched off, and the other one, which generates the higher frequency movements for swimming, takes over. Salamander robotica does exactly this – when the frequency of the walking gate reaches a certain frequency, it becomes swimming gait. Its gait can switch between crawling to walking or swimming, simply by changing the frequency of the movements.
The model developed by Ijspeert’s group suggests that neural mechanisms within the spinal cord can modulate the velocity, direction and type of gait. The findings, which are relevant to all tetrapods, shed light on the transition from aquatic to terrestrial locomotion. If all that was required in was an increase in the frequency of neural activity in the spinal cord circuits, and a slight modification of the pattern so that it became asymmetrical, the transition from water to land would not have required a dramatic increase in the complexity of neural processing.
Ijspeert, A. J., et al (2007). From swimming to walking with a salamander robot driven by a spinal cord model. Science 315: 1416-1420. [Abstract]
Ijspeert, A. J., et al (2005). Simulation and robotics studies of salamander locomotion. Neuroinformatics 3: 171-196. [Full text]