Low power CMOS electronic central pattern generator design for a biomimetic underwater robot

Young-Jun Lee, Jihyun Lee, Kyung Ki Kim, Yong-Bin Kim, Joseph AyersNeurocomputing 2007 · DOI

TL;DR

This paper presents a low-power CMOS electronic design for a Central Pattern Generator (CPG) that mimics biological neural circuits to control the walking motion of a biomimetic underwater robot. The design achieves significant power savings by utilizing subthreshold CMOS techniques and demonstrates adaptive walking capabilities in simulations.

What Problem Does This Paper Address?

Contemporary robots often struggle with adaptability in unpredictable environments, relying on rigid algorithmic control systems. This paper addresses the challenge of creating more flexible and adaptive autonomous robots by drawing inspiration from biological systems. Specifically, it focuses on emulating Central Pattern Generators (CPGs), the neural circuits responsible for innate rhythmic behaviors like locomotion in animals. The goal is to design an electronic CPG that can generate adaptive walking patterns for an underwater robot, overcoming the limitations of traditional digital control.

Methods

The core of the proposed solution is an electronic Central Pattern Generator (CPG) designed using analog circuits. The CPG is built upon a model of electronic neurons that emulate the Hindmarsh-Rose neuron dynamics, a simplified yet effective model of biological neuron behavior. These neurons are interconnected via electronic synapses that mimic chemical synaptic transmission.

Key methodological aspects include:

  • Subthreshold CMOS Design: To achieve low power consumption, the entire circuit is designed to operate in the subthreshold region of CMOS transistors. This allows for significantly reduced power draw, crucial for battery-powered autonomous systems.
  • Hindmarsh-Rose Neuron Model: The electronic neurons are based on the Hindmarsh-Rose equations, which describe the dynamics of neuronal bursting. This model is chosen for its relative simplicity compared to more complex biological models, making it more amenable to VLSI implementation.
  • Electronic Synapse Circuits: Both excitatory and inhibitory synapses are implemented to create the necessary neural network connectivity. These circuits are designed to mimic the behavior of biological synapses, modulating the signals between neurons.
  • Modular Design: The CPG is broken down into functional sub-blocks, including a neuronal oscillator, motor synergy neurons, and command input processing. This modular approach facilitates design and analysis.
  • Adaptive Control: The design incorporates mechanisms for receiving command inputs (e.g., forward, backward, leading, trailing) that modulate the CPG's output, enabling adaptive control of walking direction and gait.

Datasets

Not discussed in the paper. The paper focuses on circuit design and simulation rather than training on large datasets.

Key Insights and Ideas

  • Biomimicry for Adaptability: The paper strongly advocates for using biological neural system principles (CPGs) as a foundation for creating more adaptive robots, particularly in unstructured environments.
  • Subthreshold CMOS for Low Power: The successful implementation of a functional CPG using subthreshold CMOS techniques demonstrates a viable path towards ultra-low-power analog neuromorphic systems.
  • Hindmarsh-Rose Model Suitability: The Hindmarsh-Rose model, despite its simplification, proves effective for generating realistic neuronal bursting and rhythmic patterns suitable for motor control.
  • Parametric Control of CPGs: The design shows that by adjusting parameters (like bias current) and providing specific command inputs, the CPG can generate a variety of adaptive motor programs, mimicking natural locomotion.

Key Results

The paper reports the following key results from HSPICE simulations:

  • Power Consumption: The simulated power consumption of the electronic CPG is 4.8mW.
  • Chip Area: The die size, including I/O pads, is 2.2mm x 2.2mm.
  • Adaptive Walking: Simulation results demonstrate that the proposed CPG design can generate adaptive walking motor programs, allowing the robot to switch between different walking directions in response to command inputs. The paper shows examples of how different command combinations lead to distinct gait patterns.
  • Neuron Output Scaling: The electronic neuron circuits were scaled to operate within the low supply voltage (2V) and to match the frequency characteristics of the biological neuron model.

Limitations and Future Work

  • Simulation-Based: The results presented are based on simulations (HSPICE and Matlab). While promising, actual hardware implementation and testing on a physical robot are necessary for full validation.
  • Simplified Neuron Model: While the Hindmarsh-Rose model is effective, it is a simplification of biological neurons. More complex neuron models might offer even greater fidelity but at the cost of increased circuit complexity and power.
  • Limited Sensory Feedback: The current design focuses on CPG generation and command input. Integrating and utilizing sensory feedback from the robot's environment to further enhance adaptive control is a natural next step.
  • Future Work: The authors suggest that this CPG circuit with a coordinate controller and command neuron is viable for building adaptive analog controllers for biomimetic underwater robots. Future work would likely involve integrating this CPG with the robot's physical plant and exploring more sophisticated sensory integration for real-world deployment.

Summary updated

BibTeX
@article{lee2007low,
  author = {Young-Jun Lee and Jihyun Lee and Kyung Ki Kim and Yong-Bin Kim and Joseph Ayers},
  title = {{Low power CMOS electronic central pattern generator design for a biomimetic underwater robot}},
  journal = {Neurocomputing},
  year = {2007},
  doi = {10.1016/j.neucom.2006.12.013},
}
Figure