Robophysical Modeling of Soft Limbless Locomotors
301
ability makes limbless locomotion an attractive strategy for robots, especially
those intended for tasks like search and rescue where the surroundings can be
confined or unstable. However, this mode of locomotion requires coordinating a
high degree-of-freedom body, a task which is further complicated by the addition
of environmental heterogeneities.
There are two broad classes for the treatment of terrain heterogeneities in
snake robots; obstacle avoidance [6] and obstacle-aided locomotion [28]. A challenge to both classes is obstacles that cause forces resisting forward motion.
Traditional snake robots have one motor per joint ([7,16,20], Fig. 1A), and thus
must use sensing and control strategies to avoid deleterious terrain interactions.
Ensuring useful robot coordination while bypassing obstacles generally
requires sophisticated control. For example, Transeth et al. [28] developed a
hybrid model to determine joint trajectories for obstacle aided locomotion. Computation can be simplified in some cases by using a lower-dimensional intermediary between motion planning and control such as shape functions [29] or virtual
functional segments [21]. Decentralized control, inspired by biological central pattern generators [9], can further reduce central control complexity by offloading
computation to local reflexes, whether to avoid obstacles [30], adapt to changes
in the environment [10,24], or use obstacles for propulsion [13,14]. These strategies make use of knowledge of the terrain in a closed-loop control. Robots can
collect this information using vision [18], contact sensing [2,27,30], contact force
sensing [15], or joint-torque measurements [29].
In contrast to most limbless robots, snakes articulate their joints using bilateral musculature. The lateral body bends the animals use to generate propulsion
during terrestrial locomotion are achieved by alternating unilateral activation of
the epaxial muscles (Fig. 1B) [12].
We previously studied the desert-dwelling Shovel-nosed snake, Chionactis
occipitalis (Fig. 1C), which uses a stereotyped serpenoid waveform (sinusoidally
varying curvature) to move quickly (but non-inertially [25]) across its natural
habitat consisting of a sandy substrate and sparse heterogeneities like rocks,
twigs, and plants [17,25]. We found evidence that this snake used a control
strategy in which it targeted the muscle activation pattern for a waveform that
allowed fast motion on the granular substrate and did not change this pattern
in response to collisions with the surroundings. Our study suggested that this
“open-loop” movement was facilitated by unilaterally activated muscles which
allowed the body to be passively deformed by the surroundings (Fig. 1D, [26]).
This work indicated that a similar unilateral activation scheme could aid snakelike robots in navigating obstacles without the need for sensing or control which
responds to the surroundings.
Addition of mechanical compliance can help prevent robots from becoming
jammed in obstacles. This has typically been achieved by adding a torsional
spring element to the actuators [20,22,24]. As these robots are still driven by a
single actuator per joint, however, directional compliance such as observed in the
animal can only be achieved using active feedback. Further, when deformed away
301
ability makes limbless locomotion an attractive strategy for robots, especially
those intended for tasks like search and rescue where the surroundings can be
confined or unstable. However, this mode of locomotion requires coordinating a
high degree-of-freedom body, a task which is further complicated by the addition
of environmental heterogeneities.
There are two broad classes for the treatment of terrain heterogeneities in
snake robots; obstacle avoidance [6] and obstacle-aided locomotion [28]. A challenge to both classes is obstacles that cause forces resisting forward motion.
Traditional snake robots have one motor per joint ([7,16,20], Fig. 1A), and thus
must use sensing and control strategies to avoid deleterious terrain interactions.
Ensuring useful robot coordination while bypassing obstacles generally
requires sophisticated control. For example, Transeth et al. [28] developed a
hybrid model to determine joint trajectories for obstacle aided locomotion. Computation can be simplified in some cases by using a lower-dimensional intermediary between motion planning and control such as shape functions [29] or virtual
functional segments [21]. Decentralized control, inspired by biological central pattern generators [9], can further reduce central control complexity by offloading
computation to local reflexes, whether to avoid obstacles [30], adapt to changes
in the environment [10,24], or use obstacles for propulsion [13,14]. These strategies make use of knowledge of the terrain in a closed-loop control. Robots can
collect this information using vision [18], contact sensing [2,27,30], contact force
sensing [15], or joint-torque measurements [29].
In contrast to most limbless robots, snakes articulate their joints using bilateral musculature. The lateral body bends the animals use to generate propulsion
during terrestrial locomotion are achieved by alternating unilateral activation of
the epaxial muscles (Fig. 1B) [12].
We previously studied the desert-dwelling Shovel-nosed snake, Chionactis
occipitalis (Fig. 1C), which uses a stereotyped serpenoid waveform (sinusoidally
varying curvature) to move quickly (but non-inertially [25]) across its natural
habitat consisting of a sandy substrate and sparse heterogeneities like rocks,
twigs, and plants [17,25]. We found evidence that this snake used a control
strategy in which it targeted the muscle activation pattern for a waveform that
allowed fast motion on the granular substrate and did not change this pattern
in response to collisions with the surroundings. Our study suggested that this
“open-loop” movement was facilitated by unilaterally activated muscles which
allowed the body to be passively deformed by the surroundings (Fig. 1D, [26]).
This work indicated that a similar unilateral activation scheme could aid snakelike robots in navigating obstacles without the need for sensing or control which
responds to the surroundings.
Addition of mechanical compliance can help prevent robots from becoming
jammed in obstacles. This has typically been achieved by adding a torsional
spring element to the actuators [20,22,24]. As these robots are still driven by a
single actuator per joint, however, directional compliance such as observed in the
animal can only be achieved using active feedback. Further, when deformed away