Design of controllers with distributed central pattern generator architecture for adaptive oscillations

This article introduces a systematic method for designing a distributed nonlinear controller to achieve multiple distinct gaits, each of which is characterized by a prescribed oscillation profile and velocity, for a class of locomotion systems. We base the controller on the central pattern generator (CPG), a neural circuit which governs repetitive motions, such as walking and swimming, in most animals. First, we establish a general method for designing a nonlinear CPG-inspired controller to assign a single gait for a linear plant; we show that this problem reduces to an eigenstructure assignment problem for which a solution has recently become available. We then extend the design to an adaptive, structured controller that can adjust the gait in response to variations in the environment. The essential problem becomes a controller design to satisfy different eigenstructure conditions for different plants; a computationally tractable formulation is provided for this problem. We provide two numerical examples using a link-chain model as a plant representative of animals that move through undulatory motions, such as leeches or eels, to demonstrate the efficacy of this theory. In the first example, we employ an analytical condition for eigenstructure assignment to design an unstructured controller that assigns a single gait for the link-chain model. The second example searches for a structured controller to assign two different gaits that can be switched using a higher level command.

Automated summary

This article introduces a systematic method for designing a distributed nonlinear controller based on the central pattern generator (CPG) to achieve multiple distinct gaits. By solving eigenstructure assignment problems, the design method allows for adjusting gaits in response to variations in the environment, with numerical examples demonstrating the effectiveness of the approach.