This study investigates how the brain learns to control complex movements through the adaptive combination of motor primitives, which are computational elements in the sensorimotor map that transform desired limb trajectories into motor commands. The research shows that humans learn the dynamics of reaching movements by flexibly combining motor primitives with Gaussian-like tuning functions that encode hand velocity. These primitives have wide tuning, which limits the brain's ability to represent viscous dynamics. The study also finds that the brain builds an internal model (IM) to approximate external forces, which is crucial for learning and adapting to novel force fields. The IM is modeled as a sensorimotor map that transforms desired arm trajectories into muscle forces through a flexible combination of primitives. The study demonstrates that errors experienced in one movement affect subsequent movements, revealing characteristics of the primitives used to generate motor commands. The mathematical properties of the derived primitives resemble the tuning curves of Purkinje cells in the cerebellum, suggesting that these cells may encode the primitives underlying learning of dynamics. The study also investigates the shape of primitives underlying the formation of the IM by analyzing the temporal dynamics of movement errors. It finds that the sensitivity of the IM to errors experienced in one target direction affects the IM for other directions. The results suggest that the brain uses basis functions with specific regions of preferred activity to learn dynamics, rather than global representations. The study further shows that the width of the Gaussian primitives influences how force estimation generalizes across directions and speeds. Wide Gaussians produce S-shaped movements, while narrow Gaussians do not. The study also demonstrates that the brain's ability to adapt to high spatial frequency force fields is limited, which aligns with recent findings that humans demonstrate a lesser ability to adapt in higher spatial frequency force fields. The study concludes that the brain composes motor commands with computational elements that are broadly tuned in arm velocity. The findings suggest a link between patterns of generalization and firing properties of cells in the cerebellum, and that the preferred direction of motor cortical cells rotates during learning of force fields due to changing input from the cerebellum.This study investigates how the brain learns to control complex movements through the adaptive combination of motor primitives, which are computational elements in the sensorimotor map that transform desired limb trajectories into motor commands. The research shows that humans learn the dynamics of reaching movements by flexibly combining motor primitives with Gaussian-like tuning functions that encode hand velocity. These primitives have wide tuning, which limits the brain's ability to represent viscous dynamics. The study also finds that the brain builds an internal model (IM) to approximate external forces, which is crucial for learning and adapting to novel force fields. The IM is modeled as a sensorimotor map that transforms desired arm trajectories into muscle forces through a flexible combination of primitives. The study demonstrates that errors experienced in one movement affect subsequent movements, revealing characteristics of the primitives used to generate motor commands. The mathematical properties of the derived primitives resemble the tuning curves of Purkinje cells in the cerebellum, suggesting that these cells may encode the primitives underlying learning of dynamics. The study also investigates the shape of primitives underlying the formation of the IM by analyzing the temporal dynamics of movement errors. It finds that the sensitivity of the IM to errors experienced in one target direction affects the IM for other directions. The results suggest that the brain uses basis functions with specific regions of preferred activity to learn dynamics, rather than global representations. The study further shows that the width of the Gaussian primitives influences how force estimation generalizes across directions and speeds. Wide Gaussians produce S-shaped movements, while narrow Gaussians do not. The study also demonstrates that the brain's ability to adapt to high spatial frequency force fields is limited, which aligns with recent findings that humans demonstrate a lesser ability to adapt in higher spatial frequency force fields. The study concludes that the brain composes motor commands with computational elements that are broadly tuned in arm velocity. The findings suggest a link between patterns of generalization and firing properties of cells in the cerebellum, and that the preferred direction of motor cortical cells rotates during learning of force fields due to changing input from the cerebellum.