Integrated Control of Biological and Mechanical Power for Standing Balance and Gait Stability after Paralysis
pdf
Abstract: This project is addressing how cyber physical walking systems (CPWS) can be designed to be safe, secure, and resilient despite a variety of unanticipated disturbances and how real-time dynamic control and behavior adaptation can be achieved in a diversity of environments. Specifically, a CPWS is being developed that seamlessly integrates: (1) a person who has a spinal cord injury (SCI) with intact and excitable lower motor nerves; (2) an exoskeleton with controllably locked/unlocked and/or passively damped joints; (3) DC motors for need-dependent joint power assistance; and (4) computational algorithms that continuously and automatically learn to improve standing and walking stability. In this “muscles first” approach, functional neural stimulation (FNS) provides most of the joint torques for walking and for maximum health benefits and, thus, as-needed assistive joint motors may be small and lightweight. The specific aims are 1) Assist the user’s muscles on an as-needed basis and for high-bandwidth stability control by adding small, low passive-resistance motor/transmission pairs to our CPWS; 2) Develop computational algorithms for system estimation, machine learning and stability control for SCI users standing and walking with a CPWS while minimizing upper extremity effort; 3) Verify system performance with able-bodied individuals and assess upper extremity reduction and balance control in individuals with SCI using the CPWS for standing and ambulation.
The exoskeleton portion of the CPWS has been designed with modularity in mind – each of the joints feature universal connectors on the proximal and distal ends, allowing for modules to be swapped with little difficulty – this allows parallel development of multiple actuators of different capabilities. The first generation actuators feature relatively large transmission ratios (100 – 157:1) achieved primarily through harmonic drives. The gearboxes have been optimized for maximum efficiency and minimum friction. Additionally, feedforward controllers have been developed and implemented that can compensate for friction, inertia, and gravity to reduce the impact of the exoskeleton on the human user. One and two stage planetary gearboxes have been designed for bench testing and evaluation. These promise much greater efficiency than harmonic drives.
We are developing sensor fusion algorithms to estimate whole body center of mass (CoM) position
(CoMP) and acceleration (CoMA) using only body-mounted inertial measurement units (IMUs). These measures will enable estimates of standing and walking stability parameters such as zero moment point (ZMP). One able-bodied subject stood and walked within the work volume of a 16-camera Vicon motion capture system (Oxford Metrics, UK) with eight Xsens wireless IMUs (Xsens Technologies, BV, Enschede the Netherlands) attached to the body at the chest (sternum), back, flanks, and posterior and anterior thighs. Retroreflective markers were affixed to the body in accordance with the Vicon Plug-in Gait marker set. Coordinates of CoMP were computed analytically from the marker coordinates and segmental mass properties estimated using the regression equations of the US Army anthropometric studies. CoMA was obtained by differentiating the CoMP data twice using a Savistky-Golay filter. A Kalman Filter algorithm was used to estimate the CoMP and CoMA from the sensors and the results were compared with the gold standard computed using the markers. We found that CoMP and CoMA can be accurately estimated from body mounted IMUs with a suitably trained Kalman filter and sensor fusion algorithm.
We developed a stabilizing controller for sagittal-plane bipedal walking and demonstrated it in simulation. At the core of the controller is a deep deterministic policy gradients (DDPG) neural network that is trained in GAZEBO, a physics simulator, to predict the ideal foot location to maintain stable walking despite external impulse loads. Additional controllers for pelvis joint movement during stance phase, and ankle joint torque during toe-off, help to stabilize the biped during walking. The simulated biped can walk at a steady pace of approximately 1 m/s, and it can maintain stability during nominal locomotion despite a 30 N-s impulse (300 Newtons for 0.1 second) applied to the torso from the back or 40 N-s from the front. The walking speed increases/decreases temporarily and returns to normal in less than 5 seconds. The network increases step length and decreases step duration accordingly to counter the impulse. When a rearward impulse is applied, the DDPG network shortens the step length with a longer step duration to adapt to a slower speed. The controller was also found to be robust to the addition of a 10kg backpack on the biped.
Explanation of Demonstration: We will demonstrate our current exoskeleton that is a part of our cyber physical walking system. It includes modular knee and hip joints that utilize backdrivable, lightweight electric motors and harmonic drive transmissions. The ankle joints are currently spring loaded, but will soon be motorized for stability control. An able-bodied person will demonstrate the system.
Submitted by Roger Quinn
on
Abstract: This project is addressing how cyber physical walking systems (CPWS) can be designed to be safe, secure, and resilient despite a variety of unanticipated disturbances and how real-time dynamic control and behavior adaptation can be achieved in a diversity of environments. Specifically, a CPWS is being developed that seamlessly integrates: (1) a person who has a spinal cord injury (SCI) with intact and excitable lower motor nerves; (2) an exoskeleton with controllably locked/unlocked and/or passively damped joints; (3) DC motors for need-dependent joint power assistance; and (4) computational algorithms that continuously and automatically learn to improve standing and walking stability. In this “muscles first” approach, functional neural stimulation (FNS) provides most of the joint torques for walking and for maximum health benefits and, thus, as-needed assistive joint motors may be small and lightweight. The specific aims are 1) Assist the user’s muscles on an as-needed basis and for high-bandwidth stability control by adding small, low passive-resistance motor/transmission pairs to our CPWS; 2) Develop computational algorithms for system estimation, machine learning and stability control for SCI users standing and walking with a CPWS while minimizing upper extremity effort; 3) Verify system performance with able-bodied individuals and assess upper extremity reduction and balance control in individuals with SCI using the CPWS for standing and ambulation.
The exoskeleton portion of the CPWS has been designed with modularity in mind – each of the joints feature universal connectors on the proximal and distal ends, allowing for modules to be swapped with little difficulty – this allows parallel development of multiple actuators of different capabilities. The first generation actuators feature relatively large transmission ratios (100 – 157:1) achieved primarily through harmonic drives. The gearboxes have been optimized for maximum efficiency and minimum friction. Additionally, feedforward controllers have been developed and implemented that can compensate for friction, inertia, and gravity to reduce the impact of the exoskeleton on the human user. One and two stage planetary gearboxes have been designed for bench testing and evaluation. These promise much greater efficiency than harmonic drives.
We are developing sensor fusion algorithms to estimate whole body center of mass (CoM) position
(CoMP) and acceleration (CoMA) using only body-mounted inertial measurement units (IMUs). These measures will enable estimates of standing and walking stability parameters such as zero moment point (ZMP). One able-bodied subject stood and walked within the work volume of a 16-camera Vicon motion capture system (Oxford Metrics, UK) with eight Xsens wireless IMUs (Xsens Technologies, BV, Enschede the Netherlands) attached to the body at the chest (sternum), back, flanks, and posterior and anterior thighs. Retroreflective markers were affixed to the body in accordance with the Vicon Plug-in Gait marker set. Coordinates of CoMP were computed analytically from the marker coordinates and segmental mass properties estimated using the regression equations of the US Army anthropometric studies. CoMA was obtained by differentiating the CoMP data twice using a Savistky-Golay filter. A Kalman Filter algorithm was used to estimate the CoMP and CoMA from the sensors and the results were compared with the gold standard computed using the markers. We found that CoMP and CoMA can be accurately estimated from body mounted IMUs with a suitably trained Kalman filter and sensor fusion algorithm.
We developed a stabilizing controller for sagittal-plane bipedal walking and demonstrated it in simulation. At the core of the controller is a deep deterministic policy gradients (DDPG) neural network that is trained in GAZEBO, a physics simulator, to predict the ideal foot location to maintain stable walking despite external impulse loads. Additional controllers for pelvis joint movement during stance phase, and ankle joint torque during toe-off, help to stabilize the biped during walking. The simulated biped can walk at a steady pace of approximately 1 m/s, and it can maintain stability during nominal locomotion despite a 30 N-s impulse (300 Newtons for 0.1 second) applied to the torso from the back or 40 N-s from the front. The walking speed increases/decreases temporarily and returns to normal in less than 5 seconds. The network increases step length and decreases step duration accordingly to counter the impulse. When a rearward impulse is applied, the DDPG network shortens the step length with a longer step duration to adapt to a slower speed. The controller was also found to be robust to the addition of a 10kg backpack on the biped.
Explanation of Demonstration: We will demonstrate our current exoskeleton that is a part of our cyber physical walking system. It includes modular knee and hip joints that utilize backdrivable, lightweight electric motors and harmonic drive transmissions. The ankle joints are currently spring loaded, but will soon be motorized for stability control. An able-bodied person will demonstrate the system.