Abstract: The ability of autonomous navigation is one of the key features supporting autonomous operation of the new generation navigation constellations. Autonomous navigation based on relative position measurements among satellites can be described as a high order nonlinear estimation problem, solving such a problem will be a big challenge for onboard computers. A new solution was introduced to improve the computational efficiency of autonomous navigation: an autonomous navigation constellation made up with three satellites was designed to reduce the difficulties in algorithm implementation. The measurements are relative satellite positions in inertial space. Square root version unscented Kalman filter( SRUKF) was introduced to deal with the nonlinearity of the relative measurements and orbit dynamics model. A new autonomous navigation algorithm was designed so that all the three satellites- positions can be determined at the same time. Simulation results show that an accuracy requirement of 5m user range error(URE) within 100 days can be met.