<p>Recent methods have achieved impressive results in generating full-body motion using only head and hand tracking data. However, these approaches still suffer from common artifacts such as knee and foot jitter, foot skating, and ground penetration. These issues arise because head-mounted devices (HMDs) do not capture lower-body data, requiring the lower-body motion to be inferred solely from upper-body inputs. To address these challenges, we present PhysAvatar, which combines neural kinematics regression and physically plausible pose generation to generate full-body motions from sparse tracking. The kinematic module is a neural network that regresses the full-body motion. The physics module is a pose correction that detects foot-skating artifacts and corrects the legs using the motion-captured clips and an optimization that refines the motion to satisfy the physical constraints while reproducing the corrected reference pose. Experiments demonstrate a clear improvement over the state-of-the-art regarding minimal motion artifacts and physical plausibility.</p>