IEEE Robotics & Automation Magazine - September 2016 - 110

neurons [16]. It is controlled with only three parameters set
by the brain: two for the direction of the base of the arm and
the third for scaling the propagation velocity profile of the
bend along the arm [17]. Food is delivered to the mouth by
creating three arm segments: the distal one is passively used
as a hand, while the other two, according to electromyography studies, are given by the collision of two stiffening waves,
one starting at the contact point and controlled peripherally
and the other starting from the arm base and controlled by
the brain. An articulated structure is then created, and the
fetching movement is obtained similarly to how it would be in
a rigid arm with an elbow. The brain thus controls two
parameters, the trigger of the stiffening wave and the elbow
angle [18]. There is a strict relation between the octopus body
and its behavior, and the development of its nervous system
provides evidence of its embodied intelligence. The octopus
lacks a central representation of the arms, and the peripheral
nervous system is especially well developed in terms of neuron number, showing an organization that fits the octopus'
special embodiment [7]. The octopus model provides numerous insights into morphological computation in soft robots.
An octopus-like, eight-arm soft robot was built by implementing such principles. It can move in water, bend and
stretch its arms, grasp objects, and reach a target [17], [19]
(see Figure 2). Building a robot arm with the same density
and similar morphology as an octopus arm facilitates the construction of reaching movements in water, which are very
efficient in terms of control and energy [20]. A proof of concept with a passive arm, made of silicone with the same density, the same iperelastic behavior, and the same conic
morphology as an octopus arm showed that in water, acceleration at the base of the arm generates a bending wave that
propagates from the base to the tip of the arm, which supports the hypothesis that the bending wave in the animal's
reaching movement is given in part by the physical interaction of the arm with water. The stiffening of the proximal part

Outside
9)

10)

On Board

Locomotion
Arms

Morphological Computation
in the Octopus Suckers
Analogous to its body and arms, the octopus sucker is a muscular hydrostat structure with no rigid parts, in which muscles and connective tissue play the roles of structural elements
and the actuation system. The musculature is arranged in
radial, meridional, and circular muscular fibers that provide
skeletal-like support and force for movement [25]. A single
sucker consists of two general regions connected by a
constricted orifice: the infundibulum, which is the disk-like

SMA Actuation
4)

5)
Vision
System
1)

of the arm is achieved by embedding cables in the silicone
arm in such a way that the longitudinal and transverse contractions are mechanically coupled. With a silicone conic arm
actuated by embedded cables, Nakajima et al. demonstrated
that in water, the body's dynamics perform the computation
needed to control the arm and switch behavior [21]. Hauser
et al. demonstrated that body nonlinearities provide computational power and that they can be modeled with massspring systems [22].
Crawling is another complex behavior with only a few control parameters in the octopus. During crawling, each of the
two arms used to push the body forward executes a four-phase
cycle: shortening, attaching to the ground, elongation, and
detaching. In an octopus-like robot, one DoF is enough to
obtain the four phases, given the correct compliance and stiffening ability and the capability for elongation and shortening.
The mechanical structure of each locomotion arm is based on
a silicone cone with a flexible steel cable embedded centrally,
which produces shortening and stiffening at the same time; a
motor and a crack mechanism produce the four cyclical crawling phases [23] with minimal control. In swimming, the complex hydrodynamics that elicit propulsion with the pulsed jet
of the octopus mantle can be obtained with one DoF, given the
proper deformability of the material, the proper morphology,
and the proper geometry of the mantle and the funnel [24].

Arm 1
2)
3 Manipulation
4
Arms

Arm 2
3 Servometer
4 Orientation Mechanism
Passive Suckers
3)

(a)

(b)

Figure 2. (a) A scheme of the octopus-like eight-arm robot developed in the OCTOPUS project (FP7-ICT 2007.8.5, FET Proactive,
Embodied Intelligence, no. 231608); the two front arms are used for manipulation and employ shape memory alloy (SMA) springs as
actuators; the other arms are used for locomotion and are silicone cones with a steel cable embedded centrally. (b) A picture of the
octopus-like eight-arm robot. (Photo by Jennie Hills, London Science Museum.)

110

IEEE ROBOTICS & AUTOMATION MAGAZINE

September 2016

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2016