< Back to previous page

Publication

Realtime Estimation and Prediction of Kinematics, Dynamics and Events for Assistive Robotic Devices: Contributions towards Assistance-as-needed

Book - Dissertation

This PhD research focuses on the development of theories and software for assistive robot technology. The targeted audience are people that are weakened (e.g. by age, by a muscle disease, by a stroke, …) but are still able to move their limbs. The weakness however results in a decrease in mobility, which can lead to difficulties for the daily tasks. The goal of the research is to investigate how robotic devices can assist these people as needed, i.e. delivering enough assistance such that they can fluently perform desired motions, but not delivering too much assistance since this would imply that the subject relies less on his own muscles which will result in further muscle weakness. An important aspect of the research is that all calculations are done in realtime. The developed methods will be validated on a lower-limb bilateral exoskeleton, which will assist people with walking, sit-to-stand, climbing chairs,... The realtime assistance-as-needed will be based on an online computation of the capability gap. This capability gap is the difference between the required strength for a certain task (e.g. a sit-to-stance movement) and the subject's capability, i.e. the maximal deliverable joint torques by the subject. To attain the goal of the research, five objectives have been defined. Objective 1: An applied musculoskeletal model for the capability gap computation. State-of-the-art musculoskeletal models allow to compute the capabiltity gap. However, the required calculations are computationally expensive and hence not suited to perform in realtime. Hence a simplified model will be developed. The computed capability gap should be optimally matched to the robotic device's degrees of freedom and limitations. Objective 2: A statistical model for the computation of the required joint torques. In a first step a statistical model is established to describe the joint trajectories that correspond to a given motor task (e.g. gait) of healthy subjects. This model is learned from a data set of typical executions of the motor task, and is adapted online based on the available sensor measurements. In a second step, the joint trajectories are translated to joint torques, coping with missing information (e.g. ground reaction forces). Objective 3: A biomimetic model for the computation of the required joint torques. Using feedback laws, it is possible to calculate the muscle activations for a certain movement. These muscle activations can be translated simultaneously into joint trajectories and joint torques. The results of this objective will be compared with the results of objective 2. Objective 4: Online calculation of the capability gap. Using the results of objective 1,2 and 3, the subjects capability can be computed. This can be compared to the required joint torques, calculated in objective 2 and 3. The difference between these two should be delivered by the robot device. Due to biarticular muscles, assistance on joint A can result in a different behavior on joint B. This effect will be investigated. Objective 5: Controlling an assistive device according to the assistance-as-needed paradigm. A constraint-based control framework, developed at KU Leuven, will be adapted and extended to implement the assistance-as-needed paradigm. The results of objectives 1, 2,3 and 4 are validated in a proof-of-concept application involving an existing bilateral lower-limb exoskeleton designed to support subjects with weakness. The main challenges are to maintain sufficient accuracy when reducing the detailed musculoskeletal models, dealing with missing online sensor information, to comply with the stringent real-time requirements in robotic devices, and to observe sufficient generality in the approach in view of application to other motor tasks.
Publication year:2019
Accessibility:Closed