In the field of gait rehabilitation lower limb exoskeletons have received a lot of interest. An increasing number of them are revised to be adapted for post-stroke rehabilitation. These exoskeletons mostly work in complement of conventional physiotherapy in the subacute phase to practice gait training. For this gait training the reference trajectory generation is one of the main issues. This is why it usually consists in reproducing some averaged healthy patient's gait pattern. This paper's purpose is to display the online trajectory generation (OTG) algorithm developed to provide reference trajectories applied to gait-oriented tasks designed based on conventional physiotherapy. This OTG algorithm is made to reproduce trajectories similar to the ones a therapist would follow during the same tasks. In addition, experiments are presented in this paper to compare the trajectories generated with the OTG algorithm for two rehabilitation tasks with the trajectories followed by a therapist in the same conditions. During these experiments the OTG is implemented in a runtime system with a 500µs cycle time on a bench able to emulate late and early patients' interaction. These experiments results assess that the OTG can work at a 500µs cycle time to reproduce a similar trajectory as the one followed by the therapist during the two rehabilitation tasks implemented.