Object manipulation encompasses a large variety of research activities, from grasping to fine manipulation. During the past decade, the interests in robot object manipulation have developed, from basic researches broad industry profiles, from physics-based modeling to sensor-based learning, from stand-alone robot applications to human robot collaboration. The driving force behind this shift is the vision that many practical robot applications, such as assembly, including inevitable modeling uncertainties, which require the use of sensor-based approaches to design reactive skills for robots. Moreover, these tasks are still difficult to be accomplished with robots alone, and need combination of the flexibility of human workers and the productivity of robots. In order to improve the collaborative assembly capabilities and efficiency, we need to answer following questions. 1) What kind of control strategies and end-effectors are needed in robot assembly tasks? How can we achieve and improve it? 2) How to model the cooperation between human and robots so that the assembly task can be implemented in an optimal way? 3) How to represent the collaborative assembly tasks in order to leverage the power of learning to design efficient controllers? The aim is to bring researchers from both industry and academia to set the basis and define core open problems for collaborative assembly, such as planning, control, learning, design and perception.