Presentationen äger rum via Zoom. Zoom-länk:

Kandidatprojektets titel

Inverse Kinematics for Arm Pose Reconstruction in Virtual Reality


Woosok Moon


Lars Arvestad


Immersion is a key property of virtual reality technology. It has previously been shown that providing the user with a virtual avatar inside VR applications can increase their embodiment and therefore immersion. However, most consumer grade VR only includes tracking hardware for head and hands, making user reconstructions of the upper body non trivial, and accurate full body reconstructions not possible. In this paper we examine how two heuristic inverse kinematics methods, forward and backward reaching inverse kinematics, and cyclic coordinate descent, together with a pole target for the elbow, can be used for arm reconstruction using a VR headset and two VR hand controllers. Both methods are run on an arm model with three joints, and compared with respect to their computational speed and ability to replicate user arm poses. The results show that forward and backward reaching inverse kinematics is the faster method, but that both methods are fast enough for real time applications, and perform very similar in regards to pose recreation when paired with a pole target. Both methods are able to recreate observably accurate arm poses, but the result is dependent on the pole position. Without a pole, most recreations do not match the user. Based on the results, we believe both methods can be used for arm reconstruction, but suggest using the forward and backward reaching inverse kinematics method since it is faster.