Tentacle-Arm

A hyper-dexterous robotic arm for confined-environment access.

One platform for environments conventional robotics cannot reach. First application: subsea inspection and repair, integrated with observation-class ROVs.

Scroll

The Architecture

Engineered around a different physical principle.

Hyper-redundant arms are usually built with cable-driven actuation — long tendons that run from base-mounted motors to each joint. The standard design uses two to three cables per joint and an equal number of motors at the base, so the base scales with the arm and portability suffers.

Our patent-pending actuation works differently: four to eight cables drive the full system, with sequentially orthogonal joints that decouple how each segment moves. The base stays compact. The arm stays dexterous.

Above the hardware, a physics-based model with neural-network error correction gives the control system precision today — and measurable improvement with every deployment.

Engineering Pillars

Three pillars define what
the system can do.

Each is independently significant. Together they define a class of capability that does not yet exist in the field. Select a pillar to go deeper.

Profile Controllability

The robot conforms to the environment.

It does not force its way through a constrained space. It navigates it.

The Tentacle-Arm is a cable-actuated continuum manipulator. Its defining capability is profile controllability — it actively shapes its own curvature to thread through confined paths.

Conventional continuum robots produce single-curvature arcs. The Xtent architecture shapes non-constant curvature profiles — adapting body geometry to clear apertures, bends, and offset passages while holding tip control and payload.

First application: subsea inspection and intervention, integrated with observation-class ROVs. The architecture is environment-agnostic.

14 links · 200 mm base + 170 mm pitch · Ø 130 mm Rigid-link · world-frame aligned

Fig. — Profile controllability. The arm reshapes its curvature to pass staggered apertures as the base translates. Illustrative kinematics.

The Lineage

Six decades of slender robots.
Each solved one piece.

The continuum-robot field advanced one constraint at a time. Xtent is the first to hold all of it at once: slender, portable, locally dexterous, practical to control.

1968 — 1979 · Origins

Robots that bend like a living thing.

Anderson's Tensor Arm used tendon-like filaments. Hirose built the first snake robot. Many-DOF slender bodies proved buildable.

Active Cord Mechanism · discrete segments

WhoAnderson (Scripps); Hirose (Tokyo Tech)
FocusBio-inspired locomotion; tendon arms
BreakMany-DOF slender bodies are controllable

1980s — 1999 · Theory

The field gets its mathematics and its name.

WhoChirikjian & Burdick; Robinson & Davies
FocusKinematic theory; actuation taxonomy
BreakThe backbone curve; the term "continuum robot"
κ(s)

Backbone curve · continuous, not jointed

The backbone-curve model described shape with a few parameters. Robinson & Davies classified actuation, including the cable-driven class Xtent builds on.

2000s · Trunks & Surgery

Whole-body grasping and millimetre access.

Walker's OctArm grasped with its whole body. Surgical continuum robots reached millimetre scales. Constant-curvature became the standard model.

Multi-section trunk · whole-body grasp

WhoWalker (Clemson); Webster, Simaan
FocusWhole-body grasp; surgical access
BreakContinuum robots leave the lab

2010s · Commercialization

From the lab to the reactor.

WhoCMU lineage; OC Robotics; soft-robotics groups
FocusCommercialization; accurate mechanics
BreakThey ship — at the cost of a large base
30+ MOTORS

Big base = the problem

Cable-driven snake-arms inspected reactors and engines. Cosserat-rod mechanics matured. But the systems stayed bulky — many cables, many motors, a heavy base.

2015 — 2024 · A new model

Simpler math. Shapes beyond arcs.

Ashwin K. P. (IISc) modelled the arm as four-bar linkages solved by optimization. Within ~2% of Cosserat-rod, far faster. General routing yields local, variable curvature.

FOUR-BAR SPIRAL CABLE

Optimization model · local curvature

WhoAshwin K. P. & collaborators (IISc)
FocusOptimization kinematics; general routing
BreakFast first-principles model; non-constant shapes

Today · Xtent

Portable. Precise. Finally practical.

WhoXtent Innovations (Ashwin K. P. & team)
FocusPortability; local dexterity; real-time control
BreakSlender, portable, locally dexterous, practical
PORTABLE LOCAL CURL

Small base + local curvature = the solution

No multiple cables per segment — the bulky base collapses. General routing gives local, variable curvature. A fast model makes it controllable and shippable.

Every prior era solved one constraint. The Xtent Tentacle-Arm is the first of its kind — slender, portable, locally dexterous, and practical to control, all at once.

Lead · Dr. Ashwin K. P.

Perception & Sensing

It senses in two directions at once.

The arm does not reach into the dark blind. It builds a map as it goes.

Internally, distributed sensing gives continuous, millimetre-level estimation of the arm's own pose — the system always knows the exact shape it is holding.

Externally, distance sensors along the body's periphery and at the tip read the surrounding geometry. Fused and run continuously, this becomes SLAM: the arm builds a live 3-D map of an environment no one has measured, while locating itself inside it.

Internal Pose Estimation Peripheral + Tip Sensing SLAM 3-D Mapping Operator AR Overlay
14 links · 200 mm base + 170 mm pitch · Ø 130 mm · distributed range sensing Mapped points: 0 · Traversing & mapping

Fig. — Body-distributed SLAM. Range sensors along the body and tip reveal an unmeasured environment as the arm threads it. Illustrative.

Perception lead · Dr. Ashith R. Babu — SLAM systems refined through deployments at previous companies.

Control & Learning

Robust by model. Reliable by learning.

A predictable backbone, with a learned layer for the messy real world.

Controlling a hyper-redundant body in a confined space is its own problem. Xtent's control is hybrid: an accurate physics-based model provides a robust, predictable backbone, and a learning-based layer adds reliability as operating conditions change.

Every deployment feeds the model — experience compounds into performance, so the system is better at its hundredth mission than its first.

Physics-Based Model Learning-Based Enhancement Active Compliance Real-Time Obstacle Avoidance Deployment Data Flywheel
14 links · 200 mm base + 170 mm pitch · Ø 130 mm · commanded S-pose held · controlled dev 0.0mm vs open-loop 0mm

Fig. — Robust disturbance rejection. The commanded pose (dashed) holds across space, subsea and land while the open-loop ghost deforms. Illustrative.

Control lead · Dr. Akhil Gopalan

See where it goes to work.

The technology is portable, locally dexterous, and practical to control. The next question is where it deploys. Subsea inspection is the first.