IEEE Robotics & Automation Magazine - March 2023 - 73

stage and to guarantee the coordination of the human-robot
velocity in the active training stage. Comparative simulation
analyses and experimental results show that the proposed passive
and active direct switching training improves the intelligence
and security of rehabilitation robot.
INTRODUCTION
As prototypical physical human-robot interaction systems,
rehabilitation training robots have been developed that help
rehabilitees with walking disorders by tracking the training
trajectories preset by rehabilitation physiotherapists. Walking
exercises can be performed to enhance the leg muscle
strength and balance ability of rehabilitees, so that their independent
walking function can be gradually
restored. A variety of training
robots have been developed to address
the problems of aging societies, such as
gait exoskeleton robots [1] and ankle
rehabilitation robots [2]. With the help
of rehabilitation robots, rehabilitees
with walking disabilities can realize
good rehabilitation outcomes through
repeated walking exercises.
In recent years, the number of stroke
"
rehabilitees in China has increased at an
annual growth rate of 10% [3]. However,
there is a serious shortage of physiotherapists
and medical resources in China,
which increases the workload of physiotherapists
and deteriorates the effectiveness of rehabilitation.
Thus, the development of rehabilitation training robots not
only decreases the work pressure of physiotherapists but also
enables rehabilitees to receive timely rehabilitation training.
Since the leg strength of rehabilitees is weak in the early training
stage, it is often necessary to passively track the training
trajectories specified by doctors. Passive training robots have
been developed, such as lower-limb rehabilitation robots with
trajectory tracking, that are suitable for the passive training of
rehabilitees with paralyzed or weak limbs [4]. A rehabilitation
robot driven by a multifilament was designed based on
a motion model and a system elbow joint model [5]. These
robots have only a passive training mode and are only suitable
for the initial rehabilitation training of rehabilitees. With the
enhancement of the rehabilitees' motor function, these robots
cannot actively participate in training.
In fact, there are some research results on the active training
of rehabilitation training robots, such as ankle rehabilitation
robots, where the subjects are required to accomplish
a task within a predefined time by rotating the ankle joints
with the assistance of the robots based on visual or auditory
instructions [6]. Additionally, a robot-assisted treadmill-based
exoskeleton gait training program with an active
assistant protocol robot was developed. This program uses
haptic feedback to enhance resist motion rather than pointing
the upper extremity ankle toward or away from the path [7]. A
particle-filtered interface function was proposed to estimate
THE NOVELTY OF THE
ROBOT IS THAT PASSIVE
AND ACTIVE TRAINING
CAN BE DIRECTLY AND
GENTLY SWITCHED
DURING WALKING.
„
and predict the locations of the rehabilitee's lower limbs and
body, thereby realizing the natural and smooth active movement
of the Japan Advanced Institute of Science and Technology
robotic walker [8]. Active and passive hybrid training
modes are also available. In [9], an end-effector typed hybrid
walking rehabilitation robot with three modes was proposed
to help rehabilitees in different rehabilitation stages, and the
three modes were passive, assisted active, and active. Further,
a rehabilitation robot named DDgo pro was developed;
its passive mode let the robot fully guide a normal walking
pattern, and the active assisted mode performed rehabilitation
training for a long time while receiving muscle assistance
from the robot [10]. The above robots have an active training
mode or passive and active mixed
training modes. However, when the
robot is in the passive training mode,
the current motion should be stopped to
switch to the active training mode. For
rehabilitees with weak leg strength, the
upper limbs need to support their body
weight, and it is very difficult to switch
the training modes back and forth.
Additionally, many commercial rehabilitation
robots have advanced significantly,
including the walking balance
robots Erigo, Lokomat, and Andago [11]
and wearable robot Hybrid Assistive Limb
(HAL) [12]. To achieve the rehabilitation
training of the leg muscles, Erigo and
Lokomat, two robots for walking balance, adopted the training
modes of a pedal and a treadmill, respectively. Due to the fact
that walking is essentially a movement on the ground, the rehabilitation
robots mentioned above are unable to simulate actual
walking training with friction between the feet and the ground. A
technique that shows promise is gait rehabilitation training while
receiving partial bodyweight support from a robot [13]. Instead
of switching directly from passive to active training mode, the
Andago device that enables bodyweight support while walking
over ground has three training modes: rehabilitee following,
straight line, and manual. HAL can only help people with disabilities
whose lower limbs cannot regain the ability to walk using
bioelectric potential signals to move in accordance with the
wearer's intention to maintain physical function.
Rehabilitative walking robots need to track training trajectories
specified by doctors. Researchers have proposed
various tracking control methods, such as fuzzy sliding mode
control in rehabilitation environments [14], robust iterative
feedback tuning control for repetitive ankle training [15], and
neural network control of rehabilitation robots using state
and output feedback [16]. However, these methods, whether
in the passive or active training mode, ignore the impact
of uncertain environments of human-robot movements on
the tracking performance, resulting in unsatisfactory tracking
accuracy. Moreover, there are no safety constraints with
regard to the velocity of human-robot movements. When the
robot's velocity is mutated, the rehabilitee safety is seriously
MARCH 2023 IEEE ROBOTICS & AUTOMATION MAGAZINE
73
IEEE Robotics & Automation Magazine - March 2023

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - March 2023
