Prism X · Prism Rover · Prism Sim

Prism Rover in Prism Sim SIM

The Prism Rover concept model, and the simulated V10 corn field it is tested in. Every plant and weed in the 24-row survey block is generated from seed 7 and recorded in a ground-truth table, so a crop-analysis model can be scored against the real answer.

The rover

Small four-wheeled rover built from a clear tote bin, with a camera mast — **Exterior**Clear 0.9-gallon latching tote on four 13 cm ag-tread wheels. Stereo camera at the front; rear mast with two row-facing cameras, the RTK GNSS antenna, Wi-Fi antenna, status light and e-stop.

Looking down into the open tote at the circuit boards and wiring — **Inside, lid off**Upper deck: Jetson Orin Nano with heatsink and fan, and two dual motor drivers. Below: 4S battery, buck converter, GNSS receiver, IMU, fuse and terminal block. Gear motors drive the wheels through the side walls.

Part	In the model
Body	Clear polypropylene tote, 29 × 19 × 11 cm outside (typical 0.9 gal size, to be matched to the real bin), blue latches
Drive	4 × 25 mm gear motors, skid steer, 13 cm × 4.5 cm lugged tyres, 19 cm wheelbase, 3.5 cm ground clearance
Compute	Jetson Orin Nano developer kit on brass standoffs
Power	4S 14.8 V 5000 mAh LiPo, XT60, blade fuse, terminal block, buck converter for the Jetson
Sensing	Front stereo depth camera, two side cameras at 27 cm for crop rows, RTK GNSS receiver and antenna at 50 cm, IMU
Safety	E-stop wired in series with the power loop, status light

A concept model for simulation only; no part has been bought or tested. The video of the rover and the 360° field views are rendering now.

The corn field

Looking down a corn alley at robot height; rows run to a tree line on the horizon — **Row view, forward camera** Between the two middle rows at 1.25 m height, looking down the 30-inch alley. The yellowing patch is faintly visible ahead on the left. `16 mm · f/8 · focus 4 m · 250 s on the Jetson GPU`

Eye-level close-up of corn plants at the edge of the nitrogen-deficient patch — **Close-up at the stress patch** Alley camera at 0.75 m, with nitrogen-deficient plants on the right. `28 mm · f/4 · focus 2 m · 307 s`

Straight-down view of corn plants, soil and residue over the stress patch — **Nadir, scouting camera** Straight down from 2.1 m over the row nearest the patch centre. `16 mm · 210 s`

Drone-height view of the whole corn field with a yellow patch in the middle — **Overview from drone height** 11 m above the south edge; the yellow patch is centred at (−2.1 m, 13.8 m). `20 mm · 440 s`

What the ground truth records

Planting positions (24 rows × 40 m)	Count
Healthy	5,009
Nitrogen deficient, mild	100
Nitrogen deficient, severe (also stunted)	40
Gap (no plant emerged)	203
All positions	5,352

Weeds, in patches	Count
Lambsquarters	83
Waterhemp	61
Green foxtail	61

Each row gives position in metres, condition, variant, rotation, scale and height. About 57,500 more plants and a canopy stand-in fill the field to the horizon; they are scenery and are not scored.

How it was made

One script builds the field from a seed. The same seed gives the same field on the Jetson and on the server.
Plants are generated, not hand-modelled: two-ranked leaves with veins, midrib and rippled margins, a senesced first leaf, and a curled whorl.
The soil has real relief: clods, the planter's seed slots, and last year's stalks lying half-buried.
Rendered with path tracing on the Jetson Orin Nano's GPU: 1280 × 720, 512 samples, peak 61 °C and about 3.8 GB of memory.

Not realistic yet

A thin line shows where the distant canopy stand-in ends.
Stressed leaves look slightly bleached from above.
The tree line is soft and blobby.
Not built yet: per-plant outlines for training, other growth stages, and the apple orchard.

Rendered 25 September 2026 from fieldtwin/corn_field.py, seed 7. Simulated imagery; no real field is shown.