Assistive robotics aims to design physically collaborative robots which are able to help human partners with cumbersome tasks; for instance, lifting a heavy plank/guard and inserting it into a frame at the ceiling. To reduce human load-share, it is expected from the robot to perform such tasks in coordination with the human partner. Uncertainty of human behavior and complex dynamics of real-world environments pose challenging problems for robotic systems. It is crucial to employ control frameworks that allow for both motion tracking and interaction/force control. Furthermore, the framework should allow for reactive and adaptive motion planning toward human behavior. To deliver these requirements, we propose a Dynamical System-based control architecture with adaptation capabilities. Our preliminary experimentation using ARMAR-6 shows promising performances to achieve such a complex task in collaboration with human users.