Assist-as-needed (AAN) is a paradigm in rehabilitation robotics based on the fact that more active participation from human users promotes faster recovery of motor functions. Moreover, the patients and public engaged and involved in our research design stressed that in order to provide safe and patient-friendly assistance, rehabilitation robotics should be equipped with different constraints while giving minimal assistance where required. Most of the current constraint-based AAN methods are only capable of providing position or velocity constraints which limit the quality of assistance that the robotic systems could provide. In this paper, we propose a multi-objective optimization (MOO) based controller which can implement both linear and non-linear constraints to improve the quality of assistance. This MOO-based proposed controller includes not only position and velocity constraints but also a vibration constraint to subside the tremors common in rehabilitation patients. The performance of this controller is compared with a Barrier Lyapunov Function (BLF) based controller with task-space constraints in a simulation. The results indicate that the MOO-based controller behaves similarly to the BLF-based controller in terms of position constraints. It also shows that the MOO-based controller can improve the quality of assistance by constraining the velocity and subsiding the simulated tremors.