A Cortico-Cerebellar Dynamical Model of Motor Learning with Variable Decision Boundaries and Temporal Error-Driven Adaptation

Real-world motor behaviors, such as intercepting moving objects or navigating around dynamic obstacles, require the integration of flexible decision-making with precise temporal execution. However, a unified computational framework that couples these two processes across multiple timescales remains elusive. Here, we develop a brain-inspired cortico-cerebellar dynamical model that bridges decision dynamics and adaptive timing control. The model incorporates an upstream, secondscale decision module with a variable boundary to optimize the speed-accuracy trade-off, and a downstream cerebellar module that learns millisecond-scale time intervals via synaptic plasticity. The integration is mediated by dual feedback loops responsible for correcting higher-level strategies and fine-tuning cerebellar timing, respectively. Simulation of dynamic ball-hitting tasks demonstrates that the variable boundary significantly reduces decision time without compromising accuracy, while the cerebellar timing mechanism ensures robust cross-timescale stability and rapid adaptation. Critically, this dual feedback loop architecture confers robust generalisation capabilities, enabling the model to master novel task conditions not seen during training. These findings embody a synergistic mechanism of variable decision boundary and cerebellar learning, providing a conceptual validation for multi-scale motor control, offering a blueprint for developing more flexible and adaptive brain-inspired robotic systems.