A spiking neural model of arm control under gravity

Thumbnail Image
Issue Date
Journal Title
Journal ISSN
Volume Title
The fields of human motor neuroscience and robotics research are largely developing independently. One of the few interactions is the recurrent error-driven adaptive control hierarchy (REACH) model. It contains a spiking neural network of adaptive arm control that produces very similar behavioural and neural data as human reaching. The REACH model captures a few brain areas that are involved in human motor control, these are the pre-motor cortex, primary motor cortex and the cerebellum. This model is quite promising, however, the simulated arm moves over a planar field, meaning that gravity has no effect on the arm, which is not realistic. In this thesis, the arm will move in a vertical plane, so the REACH model is adapted such that the spiking neural network compensates for gravity. This is done by adding the gravity term to the control signal in the primary motor cortex, which captures the effect that gravity has on the joints, such that the arm will cancel this effect out. The results show that the REACH model is able to compensate for gravity, such that the trajectory and smoothness of the movement are similar to the original model without gravity. However, this is still not as realistic as a human arm as the muscles and tendons are not modeled. Future research could focus on adding the muscles and tendons, making the simulated arm 3D or looking into other ways that gravity could be represented in the brain and comparing the spiking data from REACH to human neural data.
Faculteit der Sociale Wetenschappen