306
P. E. Schiebel et al.
the head and no added mass did not exhibit the lifting kinematics. Our hypothesis was that the mass of the head generated torque that prevented the motors
from lifting segments, and torque from the added mass countered that of the
head.
3 Results and Discussion
We sought to discover if the open-loop robot could body-buckle like the animal
(e.g. Fig. 4A, t = 117 ms) during head-on collisions with a wall, and whether it
would re-orient to travel along the wall without using feedback. Wall following
has been studied from a neuromechanical perspective in invertebrate (cockroach)
locomotion [4], with feedback control playing a critical role in task performance.
We used a vertically oriented whiteboard as a wall (Fig. 4B). The low-friction
surface of the whiteboard simplified the system so that the robot was primarily
experienced ground contact forces and wall normal forces. The substrate was
a smooth wooden surface. While we initially explored robot locomotion on a
rubber mat (Fig. 2A) that facilitated low-slip motion, we found impurities in the
substrate (ridges for grip) could cause the waveform to deform. While slipping of
the robot was higher on the wooden surface it allowed us to more easily observe
wall-induced changes to the waveform.
The robot was initially placed with its long axis perpendicular to the wall
(Fig. 4B, t = −2 s). The robot’s position was randomly chosen between each
trial to vary the phase of the wave and position of the head when it contacted
the wall. In all cases the robot performed at least one full cycle of the waveform
before contacting the wall. The experiment would stop when the long axis of
the robot was fully parallel to the wall. The snake data was from [26]. In these
experiments the animal moved freely across the model granular substrate into a
vertical whiteboard wall.
The robot began by moving across the substrate using the serpenoid waveform (Fig. 4B, t = −2 s) similar to that used by the snake (Fig. 4A, t = −64 ms).
The robot initially deformed under external forces from the wall, with large
amplitude curves appearing on the body (Fig. 4B, t = 4.5 s). The animal also
deformed, leading to areas of high curvature (Fig. 4A, t = 117 ms).
Unlike the snake, in some robot trials we initially observed a straightening
behavior after impact (Fig. 4B, t = 0 s) and the increased amplitude became
apparent after half a cycle (Fig. 4B, t = 1–4.5 s). After this initial interaction
the front of the robot began to turn until the anterior segments were parallel
to the wall (Fig. 4B, t = 8 s). If at this point the robot was not re-oriented
sufficiently to continue along the wall, the same process would repeat one or
more times (note head turning back into the wall at Fig. 4B, t = 10 s). Once
the robot was rotated enough, it returned to its initial waveform and amplitude
(Fig. 4B, t = 13 s).
While both the snake and robot turned to travel along the wall (Fig. 4C),
this process required more gait cycles in the robot. We characterized the number
of undulations the robot underwent before it successfully turned (defined as 70%
P. E. Schiebel et al.
the head and no added mass did not exhibit the lifting kinematics. Our hypothesis was that the mass of the head generated torque that prevented the motors
from lifting segments, and torque from the added mass countered that of the
head.
3 Results and Discussion
We sought to discover if the open-loop robot could body-buckle like the animal
(e.g. Fig. 4A, t = 117 ms) during head-on collisions with a wall, and whether it
would re-orient to travel along the wall without using feedback. Wall following
has been studied from a neuromechanical perspective in invertebrate (cockroach)
locomotion [4], with feedback control playing a critical role in task performance.
We used a vertically oriented whiteboard as a wall (Fig. 4B). The low-friction
surface of the whiteboard simplified the system so that the robot was primarily
experienced ground contact forces and wall normal forces. The substrate was
a smooth wooden surface. While we initially explored robot locomotion on a
rubber mat (Fig. 2A) that facilitated low-slip motion, we found impurities in the
substrate (ridges for grip) could cause the waveform to deform. While slipping of
the robot was higher on the wooden surface it allowed us to more easily observe
wall-induced changes to the waveform.
The robot was initially placed with its long axis perpendicular to the wall
(Fig. 4B, t = −2 s). The robot’s position was randomly chosen between each
trial to vary the phase of the wave and position of the head when it contacted
the wall. In all cases the robot performed at least one full cycle of the waveform
before contacting the wall. The experiment would stop when the long axis of
the robot was fully parallel to the wall. The snake data was from [26]. In these
experiments the animal moved freely across the model granular substrate into a
vertical whiteboard wall.
The robot began by moving across the substrate using the serpenoid waveform (Fig. 4B, t = −2 s) similar to that used by the snake (Fig. 4A, t = −64 ms).
The robot initially deformed under external forces from the wall, with large
amplitude curves appearing on the body (Fig. 4B, t = 4.5 s). The animal also
deformed, leading to areas of high curvature (Fig. 4A, t = 117 ms).
Unlike the snake, in some robot trials we initially observed a straightening
behavior after impact (Fig. 4B, t = 0 s) and the increased amplitude became
apparent after half a cycle (Fig. 4B, t = 1–4.5 s). After this initial interaction
the front of the robot began to turn until the anterior segments were parallel
to the wall (Fig. 4B, t = 8 s). If at this point the robot was not re-oriented
sufficiently to continue along the wall, the same process would repeat one or
more times (note head turning back into the wall at Fig. 4B, t = 10 s). Once
the robot was rotated enough, it returned to its initial waveform and amplitude
(Fig. 4B, t = 13 s).
While both the snake and robot turned to travel along the wall (Fig. 4C),
this process required more gait cycles in the robot. We characterized the number
of undulations the robot underwent before it successfully turned (defined as 70%