Machines Without Bones

Abstract

Conventional robotic manipulators are constructed from rigid links connected by discrete joints. This architecture confers precision, analytical tractability and repeatability, and it has underpinned industrial automation for six decades. It is, however, poorly suited to unstructured environments and to interaction with delicate or variable objects. The present review surveys the alternative paradigm of compliant, continuously deformable machines, examines the principal actuation strategies, and identifies the constraints that continue to limit deployment beyond the laboratory.

1. Motivation

The limitations of rigid manipulation are most readily appreciated by considering a task that human operators perform without deliberation: the retrieval of a ripe fruit from a crate. The object is deformable, its dimensions vary within a population, its orientation is arbitrary, and the force at which damage occurs is both low and imprecisely known. A rigid gripper approaching this task must first localise the object with high accuracy, then plan a grasp, then regulate applied force through feedback. Each stage introduces error, and the consequences of error are borne entirely by the fruit.

An inherently compliant device inverts this relationship. Where the structure itself yields on contact, geometric uncertainty is accommodated mechanically rather than computationally. The gripper conforms to the object because it cannot do otherwise. Control effort is thereby reduced — in certain configurations, eliminated — and the burden of adaptation is transferred from the controller to the material.

This transfer has been described in the literature as morphological computation, and the term is instructive. It proposes that a portion of what is conventionally regarded as information processing may instead be performed by physical structure, and that the appropriate design objective is not a more capable controller but a body whose passive dynamics render the difficult part of the problem unnecessary.

2. Biological precedent

Invertebrate systems provide the principal existence proof. The cephalopod arm is frequently cited: it possesses no rigid skeletal element, exhibits effectively unlimited degrees of freedom, and achieves both delicate manipulation and considerable force through the antagonistic arrangement of longitudinal, transverse and oblique musculature acting against an incompressible hydrostatic core.

Of particular relevance is the observation that a substantial majority of the animal's neurons are distributed within the arms rather than centralised. Reaching movements appear to be generated by stereotyped waves of activation propagating distally, and can be elicited in arms surgically isolated from central control. Coordination is thus achieved without centralised planning — an architecture of considerable interest to engineers, though one that has proved difficult to translate into hardware.

3. Actuation strategies

Four approaches dominate the current literature.

Fluidic elastomer actuators comprise networks of chambers embedded within a cast elastomer, typically silicone. Pressurisation causes chamber expansion; an asymmetric wall thickness or an inextensible embedded layer converts this expansion into bending. Fabrication is inexpensive and the resulting devices are robust to impact and immersion. The principal disadvantages are the bulk of the pneumatic supply, which frequently exceeds the actuator in mass by an order of magnitude, and bandwidth limitations imposed by fluid transport.

Tendon-driven continuum structures transmit force along cables routed through a flexible backbone, with motors located remotely. Power density is superior to fluidic approaches and the motors may be positioned at the base, improving distal dynamics. Friction within the routing and the resulting hysteresis complicate open-loop control considerably.

Electroactive polymers, in particular dielectric elastomers, deform in response to applied electric fields. Response times are short and efficiency is favourable, but operating voltages in the kilovolt range present insulation and safety difficulties that have restricted application largely to research contexts.

Variable-stiffness mechanisms, including granular jamming, exploit transitions between compliant and rigid states. A membrane containing loose particulate matter conforms freely under ambient pressure; evacuation of the interstitial air locks the particles into a rigid aggregate. Grippers exploiting this principle achieve secure grasps across a wide range of geometries without object-specific planning, and have been among the first soft devices to achieve commercial deployment.

4. Outstanding difficulties

Three obstacles are consistently identified.

Modelling. Rigid manipulators are described by a finite set of joint variables, and their kinematics admit closed-form solution. A continuously deformable body possesses, in principle, infinite degrees of freedom, and its behaviour is governed by nonlinear constitutive relations that are strain-rate dependent and subject to hysteresis. Reduced-order approximations — most commonly the piecewise constant curvature assumption — permit tractable control but degrade in accuracy under external loading, which is precisely the regime of interest.

Proprioception. A rigid joint may be instrumented with an encoder. A deformable segment has no equivalent, and its configuration must be inferred from distributed strain measurement, embedded optical waveguides or externally applied vision. Sensing elements that are themselves compliant remain an active area of investigation, and the integration of such elements without compromising the mechanical properties that motivated the design is non-trivial.

Durability. Elastomers subjected to cyclic loading exhibit fatigue, and failure typically originates at interfaces between materials of differing stiffness — precisely where actuation is transmitted. Reported cycle lifetimes vary by several orders of magnitude between publications, and standardised testing protocols have not been widely adopted, which materially impedes comparison across the literature.

5. Assessment

The field has matured from demonstration towards selective application. Deployment to date has concentrated in domains where the handling of variable, deformable or fragile items has resisted conventional automation: food processing, horticultural harvesting, and the manipulation of irregular packaged goods. In such settings the relevant comparison is not against a rigid manipulator, which cannot perform the task at all, but against manual labour.

Medical application constitutes a second area of activity, motivated by the observation that a device which cannot exert damaging force by construction is subject to a materially different safety analysis than one restricted by software.

It should nonetheless be emphasised that compliance is not a general improvement. Where the task is precise, repetitive and conducted within a structured environment, rigid architectures remain superior by every relevant metric and are likely to remain so. The contribution of the soft robotics programme is properly understood not as a replacement but as an extension of the manipulable domain into regions that were previously inaccessible — and, more broadly, as a demonstration that the division of labour between body and controller is a design variable rather than a fixed constraint.