Robophysical Modeling of Soft Limbless Locomotors
309
prevented forward motion (e.g. Fig. 5B, C), the passive mechanism helped resolve
jams.
Buckling behavior, previously observed seen in the wall collision assay
(Fig. 5C), was observed when the robot was jammed between posts. At t =
49 s in Fig. 5C the robot is jammed between four posts and the robot is unable
to progress. At 54 s the force from the rightmost post buckles the body, allowing
the robot to reposition itself and resume the nominal waveform, continuing past
the top post by t = 58 s.
An emergent reversal behavior occurred when jamming deformed the waveform in a way that it did not provide sufficient propulsion. In Fig. 5C at t = 43–
44 s the robot is stuck between several posts in a configuration that disallows the
nominal waveform. At 46 s the front of the robot has moved slightly backward,
allowing the head to progress around the post as seen at 49 s.
Unlike the control schemes used to prevent jamming in traditional snake
robots [29], our robot was able to solve jams without sensing the obstacles
or changing the motor activation pattern. However, this strategy did typically
require multiple undulation cycles to become unstuck, and was not always successful. A trial was ended if for 10 undulations the robot was unable to progress.
In 2/6 trials the robot did not progress.
Future work can characterize the genesis of the unsolvable jams. This can
inform whether changes to the robot, such as increasing the number of joints
and the maximum joint angles to give the robot greater flexibility more akin to
that of the biological snake, will extend the situations where the purely passive
mechanism can contend with adverse heterogeneities. Further study can also
determine whether the robot without wheels can use the posts for propulsion as
well as in which situations a more sophisticated sensing and control program is
needed.
4 Conclusion
Here, we present a novel snake robot that relies on a two-actuator joint scheme
to model bilateral muscle activation patterns seen in snakes [12]. Our robot
is completely open loop, passively deforming and adapting to the environment
without sensing or control. By offloading the control into the mechanics of the
robot, a successful strategy in legged robots [23], our robot can capture snake-like
behavior (e.g. buckling) and navigate complex terrain.
The two-actuator-joint scheme was able to mimic snake locomotion when
freely moving while also allowing snake-like passive body buckling. We note the
wall following scheme that emerges in our snake robot requires no feedback; in
this way it complements the work of [4]. We posit that the wall following could be
aided via addition of head contact sensing with a amplitude modulated turning
scheme [1]. Further, our robot was able to traverse a multi-post array, resolving
jams with passive body buckling and emergent backing behaviors.
While our work has drawn inspiration from snakes, many animals across
different environments and length scales rely on undulatory locomotion, using
309
prevented forward motion (e.g. Fig. 5B, C), the passive mechanism helped resolve
jams.
Buckling behavior, previously observed seen in the wall collision assay
(Fig. 5C), was observed when the robot was jammed between posts. At t =
49 s in Fig. 5C the robot is jammed between four posts and the robot is unable
to progress. At 54 s the force from the rightmost post buckles the body, allowing
the robot to reposition itself and resume the nominal waveform, continuing past
the top post by t = 58 s.
An emergent reversal behavior occurred when jamming deformed the waveform in a way that it did not provide sufficient propulsion. In Fig. 5C at t = 43–
44 s the robot is stuck between several posts in a configuration that disallows the
nominal waveform. At 46 s the front of the robot has moved slightly backward,
allowing the head to progress around the post as seen at 49 s.
Unlike the control schemes used to prevent jamming in traditional snake
robots [29], our robot was able to solve jams without sensing the obstacles
or changing the motor activation pattern. However, this strategy did typically
require multiple undulation cycles to become unstuck, and was not always successful. A trial was ended if for 10 undulations the robot was unable to progress.
In 2/6 trials the robot did not progress.
Future work can characterize the genesis of the unsolvable jams. This can
inform whether changes to the robot, such as increasing the number of joints
and the maximum joint angles to give the robot greater flexibility more akin to
that of the biological snake, will extend the situations where the purely passive
mechanism can contend with adverse heterogeneities. Further study can also
determine whether the robot without wheels can use the posts for propulsion as
well as in which situations a more sophisticated sensing and control program is
needed.
4 Conclusion
Here, we present a novel snake robot that relies on a two-actuator joint scheme
to model bilateral muscle activation patterns seen in snakes [12]. Our robot
is completely open loop, passively deforming and adapting to the environment
without sensing or control. By offloading the control into the mechanics of the
robot, a successful strategy in legged robots [23], our robot can capture snake-like
behavior (e.g. buckling) and navigate complex terrain.
The two-actuator-joint scheme was able to mimic snake locomotion when
freely moving while also allowing snake-like passive body buckling. We note the
wall following scheme that emerges in our snake robot requires no feedback; in
this way it complements the work of [4]. We posit that the wall following could be
aided via addition of head contact sensing with a amplitude modulated turning
scheme [1]. Further, our robot was able to traverse a multi-post array, resolving
jams with passive body buckling and emergent backing behaviors.
While our work has drawn inspiration from snakes, many animals across
different environments and length scales rely on undulatory locomotion, using