Concept product

Herrand

Built for kerbs, rain and Mondays.

Design and visualisation study by Radwan Jameel · not a real company

Meet Herrand

A delivery robot designed by simulation.

Six wheels, a self-levelling cargo pod and a 360° lidar, on a frame grown by an optimiser.

Herrand in a light studio, front three-quarter view: a white egg-shaped cargo pod in a cast aluminium frame on six wheels, a small chrome badge on the nose
Herrand from the front
Front
Herrand from the side
Side
Herrand from above
Above
The cargo pod on its own: a white ceramic egg with an orange seam light

01 · The pod

Your order rides level.

The 40 L pod hangs on a pendulum yoke, so it self-levels over kerbs and slopes, swinging up to ±20° without touching the frame. The seam light says what it’s doing.

Payload
20 kg
Volume
40 L
Swing
±20°
The cast aluminium frame on its own: an organic lattice of tapered members

02 · The frame

Grown, not drawn.

The frame is cast A356 aluminium, shaped by an optimiser around the loads it carries. It weighs under 4 kg and holds a 20 kg order with a safety factor of 18.

Mass
3.99 kg
Safety factor
18.4
30 cm drop
SF 12.8
One wheel with its hub motor

03 · Running gear

A 120 mm kerb, in the rain.

Six hub-motor wheels on a rocker-bogie keep at least four tyres on the ground as it climbs. On a wet kerb it needs grip of 0.39, and wet concrete gives 0.55, so there is margin to spare. It turns on the spot.

Wheels
6 × r 130 mm
Motors
15 Nm class
Steering
Spins in place
The lidar unit on its own

04 · Sensing

It sees the street.

A 32-beam lidar sits on a tower rib, on rubber isolators. A raycast test against the robot’s own body shows it blocks just 2.2% of the beams. Cameras, ultrasonic sensors and a soft bumper cover the ground close in.

Lidar
32 beams, 360°
Self-occlusion
2.2%
Stop, wet
0.78 m
Herrand on a city pavement at dusk, the street around it traced in cyan lidar scan lines
What it sees. Each line is a lidar beam, raycast against the street.

Inside

Every part, in its place.

Each part slides out along the axis it mounts on, then back into place.

Exploded view of Herrand: parts separated along their mounting axes
  1. 1Lidar
  2. 2Cargo pod
  3. 3Pendulum yoke
  4. 4Cast A356 frame
  5. 5Rocker-bogie
  6. 6Bumper + ultrasonics
  7. 7Hub motor
  8. 8Wheel, r 130 mm
  9. 9Battery, 360 Wh

Explore

Take it apart yourself.

Turn it in your hand, pull it apart, or switch to what its lidar sees.

Drag to turn, pinch or scroll to zoom. The system buttons show one part of the robot at a time. Explode takes it apart and puts it back together. Lidar view shows what its sensor sees. Tap any part to see its name.

Out there

Rain or shine. Rush hour or 3 a.m.

Herrand passing park railings on a bright morning
Morning. A ball rolls out of the park.
Herrand in heavy rain on an old stone street
Rain at dusk. A paper boat keeps it company.
Herrand on a sunny waterfront beside fruit stalls
Noon. Market stalls on the waterfront.
Herrand alone on a wet promenade at night
Night. An empty promenade, and it still goes.
Herrand on cobbles in an old town
Cobbles. The bogies work hard so the pod stays level.

The making of

How it was designed.

Spec first, then maths, then a frame grown by an optimiser, then the checks. Each step is a script you can run again.

  1. 01

    Write the brief.

    EU pavements, 20 kg and about 40 L of cargo, a 120 mm kerb in the rain, under €1,700 in parts. Every later choice is scored against this list.

    Payload 20 kgKerb 120 mm, wetRange ≥ 20 kmParts ≤ €1,700Strength SF ≥ 2Clearance ≥ 15 mm
  2. 02

    Solve the layout before the looks.

    A wheel can’t climb a step as tall as its own radius. Too big, and the tyres overlap; the first layout did, by 64 mm. With a 20 mm gap between tyres, that leaves 121 to 130 mm. Herrand sits at 130.

    Chart: grip needed to climb a 120 mm kerb against wheel radius
  3. 03

    Pick the drivetrain on evidence.

    Six-wheel rocker-bogie against four tri-star clusters. The tri-star needs less grip, but jolts the cargo three times harder and has nearly three times the moving parts.

    Chart: rocker-bogie against tri-star on jolt, cost, custom parts, moving parts and grip
  4. 04

    Let physics draw the frame.

    A 3D topology optimiser got the load points and keep-outs for the wheels, then removed every gram that carried no load. After fifty iterations, a hoop around the pod and a truss to the axles were left.

    The optimiser at iterations 1, 6, 15, 30 and 50

    Side view at iterations 1 · 6 · 15 · 30 · 50

  5. 05

    Turn voxels into a casting.

    Rebuilt as tapered members, a strut lattice and bone-like joints, castable in A356 aluminium. The first sizing came out at 43 kg; the second, at 3.99 kg.

    The refined cast frame
  6. 06

    Check everything.

    Linear static FE on the real geometry, a kinematic kerb climb, collision sweeps of every moving part, and a lidar raycast against the robot’s own body.

    The frame coloured by von Mises stress from the FE run, blue low to red high
    Von Mises stress, the envelope of five load cases. Blue is low, red is high. The peak is 13 MPa, against a yield of 165 MPa.
    18.4SF on yield
    28.3SF fatigue
    12.8SF, 30 cm drop
    40 mmmin clearance on the kerb
    0collisions, full travel
    2.2%lidar self-occlusion
  7. 07

    Keep the mistakes in.

    Seven were caught by these checks, from tyres through the body to a safety factor of 78 that a hand calculation said was 8. They’re listed below, because they’re the proof the checks work.

Tech specs

The numbers.

20kgpayload, about 40 L, on a self-levelling pendulum yoke
120mmkerb, wet. Grip needed 0.39 of 0.55
21.8kmrange on a 360 Wh pack
0.78mstop from 6 km/h on wet concrete
18.4safety factor on yield, worst of five load cases
12.8safety factor in a 30 cm drop (7.3 g)
2.2%of lidar beams blocked by the robot itself
€1,625parts estimate at 100 units. Catalogue prices, not quotes

What went wrong

Seven mistakes, and what caught them.

Every one was found by a test, and every test still runs.

  1. Tyres through the bodyCaught by BVH collision tests that now run on every change.
  2. Wheels overlapping by 64 mmThe layout script never checked tyre clearance. Adding it gave layout v2.
  3. A frame through the wheelsThe optimiser didn’t know the wheels existed. Keep-out zones fixed it.
  4. A 43 kg frameA volume budget isn’t a mass target. The fix was two stages: topology first, then sizing.
  5. A safety factor of 78A shortcut said 78; a hand check said about 8. FE on the real geometry replaced the shortcut.
  6. Floating voxelsLoose islands gave nonsense deflections. Island removal fixed it.
  7. Seven geometry clashesThe bumper, side cameras, lid hinge, a bogie, the pod (twice) and the rear beam. The mechanics sims found them all, and all are fixed.

Honest limits

The honest limits.

It’s a concept study.The strength checks are linear static FE on 4 mm voxels. There has been no nonlinear, contact, casting-defect or physical testing, so this is not a validated design.
The hollow frame hasn’t been simulated yet.A wider hollow member has lower bending stress at equal mass, but thin-wall FE hasn’t been run yet.
Two clearances under my own rule.The rule is 15 mm. The pod sits 12.5 mm from the frame: accepted, because it swings its full ±20° without contact and closing the gap would cost cargo. A bogie sits at 14.8 mm: 0.2 mm short, flagged for the next revision.
Built with AI, directed by me.I worked with Claude as a collaborator: it wrote much of the code with me, and I made and can explain every design decision. My hand-coding is beginner level; my strengths are the 3D, the engineering judgement and the story.
Herrand is fictional.It is a brand made up for this portfolio piece. The film’s backgrounds are CC0 photographs from Poly Haven; the sound is CC0 and public-domain recordings from Wikimedia Commons plus my own synthesis.
Engineering notes
CheckMethodResult
Frame strengthLinear static FE, 4 mm voxels, A356-T6, five load casesSF 18.4 yield · 28.3 fatigue · 12.8 drop
Kerb climbKinematic sim over a 120 mm kerb40 mm min clearance; bogie ±16° of ±25°
CollisionsBVH sweep: bogie ±25°, pod ±20°Zero contacts
ClearancesMesh distance to frameBogie 14.8 · battery 15.2 · pod 12.5 mm (exception)
Lidar occlusionRaycast, 32 beams × 360°2.2% (limit 5%)
MotorsQuasi-static kerb torque10.8 Nm needed → 15 Nm-class hub motors
ArchitectureRocker-bogie vs tri-star, at r 140 mm before layout v21.0 g vs 3.4 g jolt; €422 vs €536; 10 vs 28 moving parts

Design-study codename: Seedpod. Tools: Python (NumPy, SciPy, Matplotlib), Blender 5.1 Cycles. Film: 360 frames, 1920×804, 2.5D photographic mattes behind full-3D foreground.

Radwan Jameel · 3D and creative engineering

I design and visualise hardware from first principles.

Let’s talkSee more work