Adaptive cerebellar spiking model embedded in the control loop: context switching and robustness against noise

Area of Science:

• Computational Neuroscience
• Robotics
• Biologically-inspired Computing

Background:

• Robotic arm control often requires adaptive mechanisms to handle environmental changes.
• Cerebellar models offer a biologically plausible framework for motor control and adaptation.
• Spiking neural networks provide a computationally efficient and biologically realistic approach.

Purpose of the Study:

• To evaluate a spiking cerebellar model's control capability for a 3-DOF robotic arm.
• To investigate the impact of different control architectures (recurrent, forward, forward&recurrent) on performance.
• To assess the model's adaptability to kinematic and dynamic perturbations and input noise.

Main Methods:

• Implementation of a spiking cerebellar model utilizing synaptic plasticity (LTP/LTD).
• Integration of the model within recurrent, forward, and combined forward&recurrent control loops.
• Simulation of a 3-DOF robotic arm with controlled perturbations and varying levels of noise in mossy fiber inputs.

Main Results:

• The spiking cerebellar model successfully adapted to altered robot dynamics and kinematics across all tested architectures.
• The combined forward&recurrent architecture demonstrated superior robustness against perturbations and input noise.
• The model provided corrective actions, leading to more accurate robotic arm movements.

Conclusions:

• Spiking cerebellar models are capable of adaptive robotic arm control.
• The forward&recurrent architecture offers enhanced robustness and adaptability compared to standalone architectures.
• Synaptic plasticity is crucial for the model's ability to cope with dynamic changes and noise.