Haptic joint actions with a human partner are among the most challenging tasks a humanoid robot is expected to achieve with robustness and safety. In order to successfully achieve such tasks in a direct closed-contact interaction or through manipulating a common object, several ingredients are necessary. On the contrary to wheel mobile robots or fixed robotic arms, humanoids are underactuated robots that can be assimilated to a hybrid system. Indeed, they move by sequencing contact stances and alternate continuous motion and discreet decision on contact formation/breaking. In a human-in-the-loop haptic task setup (e.g. a humanoid robot transporting jointly a beam with a human partner), the human partner may impose instantly unexpected changes in the motion, to which the humanoid robot should react appropriately. To efficiently endow a humanoid robot with proactive and robust haptic joint actions capabilities, several issues need to be addressed. First how to realize an equal control sharing in terms of initiatives and leader-follower role switching that does not affix the humanoid in a constant follower role? This question goes far beyond the narrow scope of humanoid robotics. How to map continuous haptic interaction and force cues into reactive but discreet footprint planning? How to synchronize haptic joint action tasks with locomotion (which is linked to footstep planning)? How equilibrium is defined in haptic joint action? Etc. This talk will provide hints and a feedback from basic experiences on the HRP-2 humanoid illustrating how some of these questions may be addressed and present preliminary results from human-human haptic-motion capture that we recently undertake in order to gain in understating haptic joint actions cognition and control mechanisms.