MakarovSuites
English
EnglishDeutschРусскийУкраїнськаFrançaisEspañolItalianoPolski
Contact & licences

Engineering

Soft pneumatic gripper

Pick a tomato in a sunlit glasshouse with a gripper of four translucent silicone fingers. Air through the valve fills their chambers with the time constant of an RC circuit, the pressure bends every finger with constant curvature M/EI, and where a finger meets the soft fruit the contact follows Hertz: F = 4/3 E* sqrt(R) δ^1.5, with a peak pressure that must stay below the fruit’s bruise threshold. Lift, and the fruit stays only while friction and the cradling fingers carry its weight and the lift’s acceleration. Find the smallest pressure that holds, make the skin wet and watch the tomato slip, then try a strawberry the same pressure bruises; every joule the air does is followed into the bent silicone, the dents and the losses.

Soft pneumatic gripper — Actual application image
Actual application image

What happens

Four PneuNet fingers, 96 mm long, hang from the palm at 46 mm radius; by symmetry the fruit sits on the axis and one finger is computed, the totals four times its own. A finger is eight segments of 12 mm: each joint has the bending stiffness k = EI/l = 0.1 N·m/rad (EI = 1.2×0.001 N·m²), a viscous loss c = 0.15 s·k and the chambers’ moment M = p·A·e with A·e = 0.78 cm³ - the volume each joint adds per radian, so M·dφ is the p·dV work. Without contact every joint turns M·l/EI: constant curvature κ = M/EI. The fingers are overdamped, c·dφ/dt = M − k·φ + Q. The pressure follows the valve, dp/dt = (p_s − p)/τ with τ = 0.25 s, exact over each step. The pad, 8 mm from the centre line, indents the fruit (a sphere) by δ = R + t − d; Hertz gives F = 4/3·E*·√R·δ^1.5 with 1/E* = (1 − ν_f²)/E_f + (1 − ν_s²)/E_s, the peak pressure p0 = (2E*/π)·√(δ/R) and the contact radius √(R·δ); Q is the force’s virtual work on the joints. Neighbouring fingertips press on each other once they meet near the axis. Tomato: R = 32 mm, 140 g, E = 0.45 MPa, bruise 28 kPa; cherry tomato 14 mm, 12 g, 45 kPa; strawberry 17 mm, 20 g, 0.12 MPa, 14 kPa; silicone 0.8 MPa. The arm descends, the valve opens at 1 s, at 3 s the arm lifts the fruit off its stem 15 cm, carries it 0.5 m to the crate and the valve vents. From the lift the fruit stays while 4·ΣF·(e_z + μ·|e_h|) ≥ m·(g + a_z), μ = 0.9 dry, 0.3 wet; otherwise it slips - off the fingers on the plant, or out of the hand, falling to the crate or the floor. RK4 in 0.1 ms steps, refined where a contact starts; pneumatic work = elastic + contact + dissipation + released contact energy to 10^-9 J. Newton’s method gives the static grasp, bisection on it the pressure window.

Try this

Watch the fingers close and the dents appear; read the hold margin before the lift and see the tomato carried to the crate; set the skin wet and it slips as the lift starts; press “Use the minimum” and the grip just holds; pick the strawberry and the same 32 kPa bruises it; take the fruit away and the fingertips meet.

The model and its limits

Fingers as overdamped chains of rigid segments with linear joints instead of a nonlinear silicone FEM; no finger inertia; the fruit a sphere centred on the axis, Hertz contact per segment without friction loading the fingers or slip before the threshold; the stem’s pull and the carry’s horizontal acceleration are not modelled; the bruise thresholds are illustrative. The dents are drawn four times deeper than they are.

Equations used in the module
Equations
M = p·A·e,  c·dφ_j/dt = M − k·φ_j + Q_j,  k = EI/l  (free: κ = M/EI)
dp/dt = (p_s − p)/τ,  p(t) = p_s·(1 − e^(−t/τ))
F = 4/3·E*·√R·δ^1.5,  p0 = (2E*/π)·√(δ/R),  4·ΣF·(e_z + μ·|e_h|) ≥ m·(g + a_z)
∫p·dV = ½k·Σφ² + 2/5·ΣF·δ + ∫c·Σ(dφ/dt)²·dt + E_released

Related experiments

Back to catalogue
Snake robot — Actual application image

Engineering

Snake robot

At dusk, in front of a collapsed building, a rescue team tests its snake robot on a taped field of raked dust: eight orange links on passive wheels, each joint driven along the serpenoid gait alpha sin(omega t + i beta) + gamma. Nothing pushes the robot forward but the ground - the wheels let a link slide along itself ten times more easily than across - and its translation and turning come out of the force and torque balance of the whole chain. Watch it crawl at the averaged speed the massless chain predicts, make the friction equal along and across and see it wriggle on the spot, set a small amplitude and meet the second-order closed form, offset the joints and follow it round a circle close to the ideal track. Every joule the joint motors put in reappears as heat in the dust.

Load-assist exoskeleton — Actual application image

Engineering

Load-assist exoskeleton

Lift a box off a pallet in a bright warehouse and see where the load goes. A planar mannequin - shank, thigh, pelvis, trunk, arms - follows a smooth minimum-jerk lift, and inverse dynamics gives the moment every joint must carry; at rest it is exactly the static moment of the body and box beyond the joint. The exoskeleton is a torque pair between trunk and thighs: a passive spring hands back exactly the energy it stored in the bend, an active drive takes a share of the lumbar moment, up to its torque limit and a delay late. Compare no exoskeleton, passive and active on the board, a stoop against a squat, a near box against a far one: watch the lumbar moment, the L5/S1 compression estimate, the squared-moment effort and the centre of pressure under the feet, while the human’s and the exoskeleton’s work add up exactly to the energy of body and box.

Haptic teleoperation — Actual application image

Engineering

Haptic teleoperation

Feel a remote wall through two coupled handles. Compare delay, contact stiffness and an energy-limited force-feedback channel.