Improving transparency of powered exoskeletons using force/torque sensors on the supporting cuffs

Abstract

Most of the control strategies embedded in recent robotic exoskeletons for rehabilitation and assistance are specific implementations of the well-known "assistance as needed" paradigm. A key point in the design of these systems is the requirement for the robot to exert negligible interaction forces to the wearer if he/she is performing well. Optimizing transparency of a device is a challenging task: various strategies have been proposed to achieve this goal, involving both the mechanical structure of the robot and the control algorithms. In this work, we propose a simple yet effective approach that requires minimal redesign efforts in the robotic structure and in the controller to be implemented on existing devices. We experimentally validate the method by comparing kinematic, kinetic and electromyo-graphic data collected from 3 healthy subjects as they walked in three different conditions: free treadmill walking, walking in a robotic trainer with a traditional zero-impedance configuration and walking in the same robot with the new zero-impedance configuration. Results show that the novel configuration was capable of effectively reducing the interaction forces and, as a consequence, it affected subjects' natural gait less than the traditional one did.

Muscular activity when walking in a non-anthropomorphic wearable robot

“…The symbols τ h and τ k indicate hip and knee torques respectively. It is worth noticing that torques needed to backdrive the robot were similar to the one obtained in other works where active control was indeed used to improve transparency of the robot [16]. Moreover, these values are only in the order of 10% of the torques deliverable by human joints during locomotion.…”

“…Moreover, different control strategies are often adopted to actively increase robots backdrivability, e.g. by resorting to: closed-loop control based on torque feedback from the actuated joints (zerotorque approach [14], [15]) or on torques/forces measured at the interface between robot braces and human limbs [16]; active compensation of friction and inertia [17]. Anyhow, active control approaches to increase backdrivability are affected by some issues: i) an effective action is only possible within controller bandwidth limits; ii) the on-line estimation of required position derivatives introduces noise and delay; iii) non-colocation of sensors with respect to actuators can cause instability.…”

Acceleration-based transparency control framework for wearable robots

“…Common and intuitive model-based controllers include gravity and friction compensation, which only require models from the robot and not from the human. For instance, [5] uses such compensations in addition to an interaction force feedback control. In [6], besides joint torque feedback controllers, a feedforward command that realises a conservative force field is proposed.…”

Interaction Force Estimation for Transparency Control on Wearable Robots Using a Kalman Filter

“…For instance, BLEEX uses a low impedance (high-sensitivity) controller [1] while HAL uses electromyography (EMG). More recently, other works have also investigated admittance control with positive acceleration feedback [5] and interaction force feedback with modelbased gravity compensation [3]. Our approach is based on feedforward and feedback control loops on the acceleration signals from both human and robot.…”

“…More specifically, this sensory capability may enable or enhance the robot's ability to perform tasks such as load carrying [1], trajectory tracking [2], and being imperceptible (i.e. transparent) to the user [3], [4].…”