One of the critical aspects in the design of an assistive wearable robot is the energy efficiency of the actuation system, since it affects significantly the weight and consequently the comfort of the system. Several strategies have been used in previous research, mostly based on energy harvesting, compliant elements for mechanical energy accumulation (springs or elastic cords), ratchets and clutches. However, the design of the optimal actuator arrangement is highly dependent on the task, which increases significantly the complexity of the design process. In this work we present an energy efficiency analysis and design optimization of an actuation system applied to a soft module lower limb exoskeleton. Instead of performing a comparison between predefined mechanism arrangements, we solve a full optimization problem which includes not only the mechanism parameters, but also the mechanism architecture itself. The optimization is performed for a walking task using gait data from a stroke subject, and the result is a set of actuator arrangements with optimal parameters for the analyzed task and selected user. The optimized mechanism is able to reduce the energy requirements by 20-65%, depending of the joint. The proposed mechanism is currently under development within the XoSoft EU project, a modular soft lower-limb exoskeleton to assist people with mobility impairments.