Kinematic chain motion predictions using results from multiple approaches combined via an artificial neural network

A system having sensor modules and a computing device. Each sensor module has an inertial measurement unit configured to track its orientation. In a kinematic chain, multiple rigid parts of a user are connected via joints. At least one rigid part is not independently tracked using sensor modules. The computing device computes the estimates of the orientation of the rigid part, separately using an artificial neural network model or using assumed orientation relations. During a sequence of actions performed by the user, orientation estimates produced by one technique can be more accurate than another technique at some time instances, but less accurate at other time instances. An artificial neural network is trained to classify the accuracy the estimates and/or to combine the estimates to provide improved orientation estimates for the duration of the sequence of actions.