I've had a better time using a stand-in entity that handles movement and pathfinding with an agreeable shape for collision, and then a visual entity that handles animation and collision with bullets/weapons. Although there is legwork to do to synchronize the 2 entities on top of simply placing the visual entity in the same spot. You can also smooth out erratic movements when entities collide into each other, by causing the visual entity to smoothly move to it's collision entity. But this is on a per game thing.

@JCL Out of curiosity, what kind of collision tasks are done at the end of every frame? Identifying and resolving boxes that collide then attempt to moving them the next frame?