In　the　kinematic　design　of　wearable　exoskeletons,　the　issue　of　axis　misalignments　between　the　human　and　the　exoskeleton　joints　should　be　well　dealt　with.　Otherwise,　large　physical　human-robot　interaction　(p-HRI)　forces　may　occur　at　the　human-robot　interfaces,　which　makes　the　p-HRI　uncomfortable　or　even　unsafe.　To　cope　with　this　issue,　a　kinematically　compatible　design　approach　of　wearable　exoskeletons　has　been　investigated　by　researchers,　and　great　development　has　been　made　in　recent　years.　Moreover,　the　influence　of　such　a　design　on　the　exoskeleton＇s　p-HRI　performance　should　be　evaluated　to　determine　if　the　design　is　feasible.　In　this　paper,　a　self-adapting　lower-limb　exoskeleton　mechanism　for　three　degrees　of　freedom　gait　training　is　proposed,　and　the　mechanical　structure　of　the　exoskeleton　mechanism　is　designed　in　detail.　Then,　based　on　the　presented　exoskeleton　mechanism　and　the　use　of　suitable　force/torque　sensors,　a　p-HRI　force　measurement　system　is　developed.　Subsequently,　the　p-HRI　forces　of　the　human-robot　closed　chain　under　the　static　and　motion　modes　are　detected,　and　the　influence　of　the　self-adapting　design　on　the　lower-limb　exoskeleton　mechanism＇s　p-HRI　force　feature　is　evaluated.　The　results　indicate　that　additional　human-robot　connective　joints　could　reduce　the　p-HRI　force　significantly,　the　compatible　design　of　the　exoskeleton　mechanism　is　effective,　and　is　thus　applied　to　human　lower-limb　gait　training.