Non-linear adaptive control inspired by neuromuscular systems.

Area of Science:

• Neuromorphic Engineering
• Computational Neuroscience
• Robotics

Background:

• Current neuromorphic computing primarily uses spiking neuron models.
• Neuro-mechanical control principles offer alternative computational paradigms.
• Muscle fiber recruitment and impulse responses provide a basis for analog process control.

Purpose of the Study:

• To propose and validate a novel neuromorphic computing model inspired by neuro-mechanical control.
• To implement a system capable of timing, output quantity representation, and wave-shape approximation for analog processes.
• To demonstrate the model's adaptivity and robustness for controlling artificial muscles.

Main Methods:

• Developed an electronic model of a single motor unit for generating muscle twitches.
• Constructed random ensembles of motor units for agonist and antagonist muscle simulation.
• Integrated a multi-state memristive system for adaptive time constant control.
• Utilized SPICE-based simulations to test control tasks including inverted pendulum, whack-a-mole, and handwriting.

Main Results:

• Successfully simulated a single motor unit's twitch generation.
• Demonstrated control of timing, amplitude, and wave shape in complex tasks.
• Validated the model's effectiveness in electric-to-electronic and electric-to-mechanical applications.
• Showcased adaptive control capabilities through memristive elements.

Conclusions:

• The proposed neuro-mechanical model offers a viable alternative to traditional neuromorphic approaches.
• Ensemble-based control with local adaptivity enhances robustness for artificial muscle applications.
• This approach holds potential for advanced control in soft robotics and bio-inspired systems.
• The model's adaptability addresses challenges like varying conditions and fatigue in artificial actuators.