Page 38 - Read Online

Page 20 of 26 Blewitt et al. Soft Sci 2024;4:13 https://dx.doi.org/10.20517/ss.2023.49

[73]
the cables pulling and causing the segments to expand, with slackness occurring on contact with the wall .
Kandhari et al. detected this by analysing load data from the cables and identifying peaks with a median
[73]
filter . This work was further developed by Kandhari et al. who added 36 force sensors to the robot and
developed a closed-loop control system where the robot could estimate its shape and environment during
movement . The robot expansion was controlled by a desired force setpoint, whilst the contraction was
[74]
controlled by stretch sensors. This resulted in the prevention of high-force interaction and reduced forward
slip improving the motion. Calderón et al. created two robots: the first being a basic pneumatic inchworm
robot and the second the same except for Stretchable Liquid Circuits (SLCs) integrated into the gripping
units . These extra features provided strain sensing, meaning they could determine when enough force was
[75]
being applied to the pipe walls. The control system for the robots can be seen in Figure 18.

The Multiple Input Multiple Output (MIMO) controller uses the same sequential on-off locomotion rules
observed in other inchworm mechanisms. In the first robot without integrated strain sensors, the actuators
are expanded/deflated until a set pressure is reached using PID control, whereupon, locomotion rules
determine the next action. In the second robot, a setpoint pressure is still used to guide the PID inflation of
the actuators, although readings from the strain sensor are used to monitor whether the robot has made
contact with the pipe surface. Calderon et al. found that when the gripping actuators are expanding without
external compression from the pipe wall, the resistance of the SLCs increases linearly, whereas, once under
[75]
compression, the behaviour is non-linear . Therefore, the actuator can be said to have made contact with
the pipe wall when the change in voltage with time no longer approximates to a constant. In the second
robot controller, the voltage from the sensors is used to indicate pipe wall contact. If detected, the pressure
supply to the actuator is stopped and the setpoint pressure is updated to the current value. Adaptive control
systems such as this allow robots to move through less predictable environments. Calderon et al. used this
control system to efficiently move the second robot through a pipeline of changing diameter .
[75]

As demonstrated by Zhang et al., the constant curvature model can be used to model how an inchworm
robot may approach a T or Y junction, although better sensing would be required for this to be considered
robust . The robot should be able to
[55]
(a) Identify its desired position in task space.
(b) Control its movement towards the goal position.

To achieve (a), the robot would be required to detect a junction and determine the relative position of the
junction relative to its body so that it can move in that direction, something that could be done using
machine vision or other environmental sensors . To tackle (b), live sensing of movement is required.
[76]
Goldoni et al. made stretchable nanocomposite sensors that could be integrated into robot design . These
[77]
were demonstrated to provide identifiable readings for bending, interaction with obstacles and stretching.
To facilitate this, a pair of sensors are placed on either side of each other on the soft actuator [Figure 19],
with the readings compared to infer the state of the actuator. This works well for a planar robot; possibly,
similar methods could be used for 3D movement.

Navigation
Sensing of the environment is vital for the facilitation of movement control, mapping and navigation. Pipe
inspection robots are most effective when robots can autonomously or semi-autonomously navigate
[78]
pipework whilst feeding back information about conditions . To do this, the position of faults or hazards
must be determined, usually with Serial Localisation and Mapping (SLAM) . Currently, a few worm robots
[76]
have successfully demonstrated SLAM. Ishikawa et al. developed an odometry method to help determine
the shape of the pipe the robot was travelling through . A sensing unit was placed at the tail of an
[79]