Control and Dynamic Manipulability of a Dual-Arm/Hand Robotic Exoskeleton System (EXO-UL8) for Rehabilitation Training in Virtual Reality
- Author(s): Shen, Yang
- Advisor(s): Rosen, Jacob
- et al.
Every year there are about 800,000 new stroke patients in the US, and many of them suffer from upper limb neuromuscular disabilities including but not limited to: weakness, spasticity and abnormal synergy. Patients usually have the potential to rehabilitate (to some extent) based on neuroplasticity, and physical therapy intervention helps accelerate the recovery. However, many patients could not afford the expensive physical therapy after the onset of stroke, and miss the opportunity to get recovered. Robot-assisted rehabilitation thus might be the solution, with the following unparalleled advantages: (1) 24/7 capability of human arm gravity compensation; (2) multi-joint movement coordination/correction, which could not be easily done by human physical therapists; (3) dual-arm training, either coupled in joint space or task space; (4) quantitative platform for giving instructions, providing assistance, exerting resistance, and collecting real-time data in kinematics, dynamics and biomechanics; (5) potential training protocol personalization; etc.
However, in the rehabilitation robotics field, there are still many open problems. I am especially interested in: (1) compliant control, in high-dimensional multi-joint coordination condition; (2) assist-as-needed (AAN) control, in quantitative model-based approach and model-free approach; (3) dual-arm training, in both symmetric and asymmetric modes; (4) system integration, e.g., virtual reality (VR) serious games and graphical user interfaces (GUIs) design and development.
Our dual-arm/hand robotic exoskeleton system, EXO-UL8, is in its 4th generation, with seven (7) arm degrees-of-freedom (DOFs) and one (1) DOF hand gripper enabling hand opening and closing on each side. While developing features on this research platform, I contributed to the robotics research field in the following aspects:
(1) I designed and developed a series of eighteen (18) serious VR games and GUIs that could be used for interactive post-stroke rehabilitation training. The VR environment, together with the exoskeleton robot, provides patients and physical therapists a quantitative rehabilitation training platform with capability in real-time human performance data collection and analysis.
(2) To provide better compliant control, my colleagues and I proposed and implemented two new admittance controllers, based on the work done by previous research group alumni. Both the hyper parameter-based and Kalman Filter-based admittance controllers have satisfactory heuristic performance, and the latter is more promising in future adaptation. Unlike many other upper-limb exoskeletons, our current system utilizes force and torque (F/T) sensors and position encoders only, no surface electromyography (sEMG) signals are used. It brings convenience to practical use, as well as technical challenges.
(3) To provide better AAN control, which is still not well understood in the academia, I worked out a redundant version of modified dynamic manipulability ellipsoid (DME) model to propose an Arm Postural Stability Index (APSI) to quantify the difficulty heterogeneity of the 3D Cartesian workspace. The theoretical framework could be used to teach the exoskeleton where and when to provide assistance, and to guide the virtual reality where to add new minimal challenges to stroke patients. To the best of my knowledge, it is also for the first time that human arm redundancy resolution was investigated when arm gravity is considered.
(4) For the first time, my colleagues and I have done a pilot study on asymmetric dual-arm training using the exoskeleton system on one (1) post-stroke patient. The exoskeleton on the healthy side could trigger assistance for that on the affected side, and validates that the current mechanism/control is eligible for asymmetric dual-arm training.
(5) Other works of mine include: activities of daily living (ADLs) data visualization for VR game difficulty design; human arm synergy modeling; dual-arm manipulation taxonomy classification (on-going work).