The arm bearing is the weak link in the precision cycloidal reducer used in robots, often determining the fatigue life of the reducer’s precision retention. To extend the fatigue life of the arm bearing, a series of studies were conducted. The theoretical fatigue life equation for the arm bearing was first obtained by deriving a generic theoretical force equation. After simulating dynamic contact between rolling elements using Hertz elastic contact, the needle roller-cage contact was similarly modelled with spring stiffness and damping. This resulted in the creation of a quasi-dynamic differential equation for the arm bearing, which was then solved using the Runge-Kutta method in conjunction with an enhanced slicing method. After that, an arm bearing optimization model was created, with the basic dynamic load rating as the optimization goal. Design variables included needle roller diameter, bearing pitch circle diameter, effective contact length of needle rollers, and number of needle rollers. A Genetic Algorithm was used to solve the problem, and the arm bearing’s fatigue life was increased. Finally, finite element models of the optimized arm bearing were created using finite element software. Stress distribution within the roller-cage assembly was analyzed, and the fatigue life was evaluated with considering roller tilt. These analyses verified the feasibility and effectiveness of the optimization results. The results indicate that optimizing the basic dynamic load rating of the arm bearing, considering needle roller tilt, effectively improves fatigue life, significantly strengthening this vulnerable aspect of the precision cycloidal reducer and providing valuable insights for practical engineering design.