12AM RoboticsTalk to us

Everything you see here is simulated.

Digital twins of real farms.

Simulated orchards for agricultural robots. Real plant structure, real physics, and labelled sensor data on every frame. Test your robot before the season starts.

Talk to usSee what it sees

A season in a day

Field robots get one harvest a year to test on. In simulation, the whole season runs in minutes and can be repeated as often as you need.

Branches grow, leaves fill in, apples ripen from green to red and drop to the ground under physics.

Apple row from bare trellis to fruit drop.

Sensor outputs

Camera, depth, segmentation and bounding boxes from the same render, aligned to the pixel. No manual labelling.

Simulated apple orchard row at golden hour with a worker in a high-visibility vest
The RGB camera view, rendered with RTX path tracing at golden hour.

Physics-enabled assets

Every stem has a detachment force. Branches bend and spring back, and leaves collide with the gripper.

Each pick is scored as a success, a miss or a damaged branch.

Apple picking with a two-finger gripper.
Pomegranate picking with a pole tool.
Greenhouse strawberry: stems bend, the berry detaches and the stem recoils.

Simulated picks

The gripper, the plant and the detachment are all simulated. Contact forces, stalk failure and the fruit's swing come from the physics, not from animation.

Three soft fingers wrap the apple, roll it and lift it. The stalk parts at the spur under combined pull and twist, and stays on the fruit.

Plant to orchard

Each crop starts as one plant with its real structure: trunk, branches, leaves and fruit. It then varies into full rows at real planting spacing.

Apple

High-density trellis orchard: 2,405 trees in 37 rows of 65.

A single simulated pomegranate bush with multiple trunks and red fruit, standing on dry soil

Pomegranate

Multi-stem bushes, 63 plants in 7 rows at 4.0 m by 2.8 m.

Greenhouse crops

Vine, truss and trough crops built with the same pipeline: stems, leaves and fruit as separate parts, ready for physics.

Illustration of a greenhouse tomato vine with green, turning and red fruit on its trusses
TomatoTrained vine with trusses. Stems and leaves flex; each fruit detaches.
Illustration of a hydroponic cherry tomato vine on a grow slab with long trusses of small fruit
Cherry tomatoHydroponic vine on rockwool with full trusses.
Illustration of a two-stem greenhouse bell pepper plant with red and green peppers
Bell pepperTwo-stem V-system on twine, fruit hanging in the forks.
Illustration of a strawberry plant with ripe berries, a flower and runners
StrawberryTrough-grown. Stems bend and the stalk can be cut.

Illustrations.

Varied environments on demand

Sun, overcast, golden hour, bare soil or grass cover. Change the conditions without changing the orchard.

Sim-to-real results

We pretrained an apple detector on simulated orchards, fine-tuned it on a few real photos, and tested it on real orchard images. We compared it with the same detector trained on real photos only.

With 7 real images, the sim-pretrained model beats a real-only model trained on 67.

Detection accuracy against number of real labelled imagesSim-pretrained: 0.376 AP at 7 images, 0.423 at 34, 0.426 at 67, 0.446 at 168, 0.449 at 670. Real only: 0.229, 0.317, 0.347, 0.397, 0.447.0.20.30.40.573467168670Real labelled images used for training (log scale)Sim-pretrainedReal photos only
Detection AP (COCO, IoU 0.5–0.95) on the public MinneApple test set, mean of three seeds. Paper in preparation.

How it works

  1. Step 1

    Model the crop

    We build the plant from its real structure and give each fruit a detachment force.

  2. Step 2

    Grow the orchard

    Plants vary and fill rows at real spacing, under the conditions you need.

  3. Step 3

    Connect your robot

    Bring your robot model. Your software connects over ROS 2, the same as on hardware.

  4. Step 4

    Run, score, export

    Run thousands of trials overnight, score them, and export labelled data.

Get in touch

Write to info@12amrobotics.com

  • Robotics teams

    Building a picker, pruner or sprayer? Test it in our simulated orchards.

    Talk to us

  • Researchers

    Plant models, physics and labelled synthetic data for perception and manipulation research.

    Propose a collaboration

  • Investors

    We're talking to investors about our first institutional round.

    Ask for the deck

  • Engineers

    Hiring across simulation, physics, rendering, perception and ML. Send us something you've built.

    Get in touch