• Semi-active
• transtibial
• prosthesis
• control
• middle level control
• motor control
• finite state machine

Passive prostheses fail to provide the same amount of energy as able-bodied individuals do when walking. Therefore, a semi-active transtibial prosthesis based on a passive ESAR prosthetic foot has been designed at the Wearable Robotics Lab at the Biorobotics Institute of Sant’Anna – School of Advanced Studies. This prosthesis efficiently collects energy during stance by using a unidirectional clutch and a motor that can impose additional dorsiflexion on the structure. The collected energy can be released at a desirable timing using a proper control method. Designing, implementing, testing, and evaluating a part of such a control method, that is the middle-level control, is the objective of this thesis.

First, a literature review on the state of the art of middle-level control methods for lower limb prostheses is conducted. Then, the mechanical design of the prosthesis is reviewed, noting relevant aspects for the control viewpoint. Subsequently, a stride segmentation method and four control methods are designed. Three of them are implemented, tested and evaluated. The stride segmentation is based on the raw inertial data of the foot. The first control method is a passive mode that actively unloads the structure during swing phase. The second and third methods actively provide an energy release during late stance. By means of adaptive thresholds, the motor can impose additional deformation to the structure and manage the timely release of the elastic energy gathered during the stance. Both active methods are adaptive to changes in gait velocity with a delay of one stride.

A dataset was created with data for three able-bodied subjects wearing the prosthesis through an adapter. The subjects walked on a treadmill at four different velocities and using the prosthesis in four different conditions: (i) the passive mode, (ii) the two active modes, and (iii) with the motor disabled.

The stance detection performed well, detecting 99.9% of the strides that were taken throughout all the trials. An analysis of the timing of the toe-off events was performed but was not feasible concerning the heel strikes. The passive mode was also successful in 99.9% of the strides taken in respective trials. The first active method delivered an energy release with a total success rate of 97.6% and the second active method delivered the energy release in all relevant strides, therefore having a success rate of 100%. The timing of the energy release, tailored on subjects’ preferences, resulted later than what was suggested by biomechanical studies. Future activities will mainly focus on an in-depth analysis of the timing of the energy release, along with a comparison of the timing of the detected heel strike and toe-off events to the real events. The overall performance of the designed control methods is promising and will be further developed and tested on amputees.