Abstract: ‍ ‍With the increasing sensor resolution capabilities （such as the phased array radar）， obtaining multiple measurements from an object body is possible； therefore， treating an object as a point mass becomes less valid as it would result in a potentially significant loss of information. In this case， the traditional methods for tracking an object cannot be applied directly. In contrast， extended object tracking algorithms， which consider not only the kinematic state （such as the position， velocity， and acceleration of the extended object） but also the extension （such as the shape， size， and orientation of the extended object）， could provide more accurate， reliable， and comprehensive estimates of the extended object’s state. Recently， using a random matrix for extended object tracking has gained popularity. In practice， the state of the extended object is generally complex. When an extended object maneuvers， both the kinematic state and the extension may undergo abrupt changes. Multiple model approaches， such as the interactive multiple model （IMM）， are well-known candidates for significantly improving overall tracking performance if the extended object switches between maneuvering and non-maneuvering behavior. In this study， we considered tracking a maneuvering extended object based on the distributed network. We proposed a distributed maneuvering extended object tracking algorithm based on the diffusion strategy， where the random matrix method is utilized to model the extension of the extended object. We expanded the IMM framework to describe the motion characteristics of the extended object with different extension characteristics under different process noises and further developed a communication-cost-effective partial diffusion strategy. Specifically， each node in the distributed network would track the maneuvering extended object based on the IMM method， where the model data would be fused using the weighted Kullback-Leibler average （KLA） method. Additionally， utilizing the partial diffusion strategy， each node would only share partial intermediate estimates with its neighbors during each iteration， which would markedly reduce the communication burden among nodes at the cost of moderate or even slight performance loss. Illustrative simulations validate that the proposed algorithm based on a partial diffusion strategy can effectively track the maneuvering extended object， incurring a relatively lower network communication burden.