header object
home
Ask Expert
header bars header close
home icon
Services down arrow down arrow active
Image Annotation Video Annotation 3D point cloud Automatic annotation Data validation Semantic segmentation Data creation Data collection Generative AI Custom AI Data compliance Keylabs: annotation platform
Industries down arrow down arrow active
Automotive Aerial Agriculture Construction Documents Fashion Livestock Logistics Manufacturing Maritime Medical Retail Robotics Security Sport Waste management
Blog Company
fb
linkedin
youtube
instagram
x

How Keymakr supported
Bluewhite in advancing
real-world autonomy

rsip vision cipia v7 AMD shopify walmart logitech Recycleye viisights jabra zkt qm qm qm

AgriTech, autonomous agricultural machinery

Download PDF
Company:
Services:
Bounding Box icon Bounding Box icon hover
Data annotation (bounding boxes, LiDAR-based annotations, segmentation)
Bounding Box icon Bounding Box icon hover
Perception tasks (lane detection, obstacle detection, segmentation)
Overview:

Intro

Bluewhite creates innovative AI-powered autonomy solutions for off-road and agricultural applications. The company's mission is to enable sustainable automation that increases productivity, empowers growers, and enhances safety, while respecting both the land and the people who work it.

At the core of Bluewhite’s approach is a modular autonomy stack, Field Intelligence AI. Designed for scalability and reliability, it allows OEMs, dealers, and growers to configure autonomous workflows to meet their specific needs. The stack supports automated driving, labor-saving tasks, and consistent, high-quality operations built to perform in real-world agricultural conditions.

While advanced perception and vision models are essential to Bluewhite’s platform, its autonomy solutions are not confined to the lab. Clients' systems are already in growers’ hands today, operating across orchards and vineyards and delivering tangible results in productivity, safety, and sustainability.

Bluewhite helps growers and off-road operators boost productivity and sustainability by automating key tasks through autonomy. The company’s Field Intelligence AI stack enables safe, scalable, and reliable operations, giving growers the confidence to do more with less while maintaining high-quality outcomes.

The challenge

Bluewhite reached out to Keymakr as a data labeling provider to help with data solutions for autonomous tractor systems used in almond orchards and soft fruit groves. The tractor needed to navigate independently, recognize obstacles, follow predefined paths, and perform smart spraying.

The Keymakr team was responsible for data annotation to enable navigation.

Key challenges of the project:

  • Gradual increase in task complexity: from simple lane detection to full environment segmentation and camera contamination analysis;
  • High labor intensity despite relatively small data volumes;
  • The model had to be trained to operate in 3D spaces using only 2D annotations. To make the process more efficient, innovative solutions were required.
Milestones and solutions
Stage 1. Lane detection and obstacle detection

At the first stage, the project focused on simple annotations: marking driving lanes and labeling obstacles such as trees and poles. Work was carried out using different datasets. The client also experimented with training the model on synthetic data generated through visual simulations (including diffusion-based approaches). However, since real-world accuracy was required, the team ultimately returned to working with real footage captured by the tractors’ onboard cameras.

Stage 2. LIDAR annotations

Bluewhite’s LiDAR data was in the form of sequences of the lidar channels (intensity, reflectivity, and range), where each pixel separately had various information, including the distance from the lidar. For annotation, this format was highly challenging: the images looked like “flat” pictures with no visual cues, making it nearly impossible to quickly understand which objects were closer to the tractor and which were farther away.

A particular challenge was that only objects within a specific range needed to be labeled. Anything beyond that was irrelevant for the model, so a custom solution was required.

Solution:

To adapt the data for 2D annotation, the Keymakr team developed a custom tool on the Keylabs platform:

  • Each depth range was assigned a distinct color (a gradient from red to green/blue).
  • The color scale was overlaid onto the LiDAR images, allowing annotators to see which objects fell into the required depth range immediately.
  • Clear rules were set: only objects of a specific color (i.e., within the defined distance) were to be labeled.

In effect, the team translated 3D information into a 2D space, making it easier and more efficient to process. This solution was created at record speed, allowing the labeling process to be completed quickly and efficiently.

Validation in Segments.ai:

Although all data annotation was performed in 2D, quality had to be confirmed in a 3D space. For this, Keymakr used Segments.ai, a partner platform specializing in 3D data workflows. Before export, results were uploaded into Segments, where the team verified how annotations mapped in 3D, checked for offsets or errors, and made adjustments where needed.

By working with Keymakr, Bluewhite accelerated the development of its perception systems while maintaining high accuracy and reliability. The 1.75–2x increase in annotation efficiency enabled Bluewhite to bring autonomy solutions to growers more quickly, helping them achieve safer and more productive operations in the field.

Stage 3. Environment segmentation

After completing the stages related to object detection and navigation, the client needed to expand the capabilities of their technology. The task was no longer limited to recognizing lanes and obstacles, but required a complete understanding of the tractor’s surroundings.

Bluewhite developed a proprietary segmentation model, while our team provided the annotated data, processing thousands of objects. Unlike the earlier LiDAR images, this stage relied on standard RGB images. Each pixel in the image had to be assigned to a specific class, such as:

  • grass
  • road (drivable / non-drivable surface)
  • trees
  • rocks
  • sky
  • other environmental elements

Result:

This milestone marked a successful transition from detection to semantic segmentation, creating more complex datasets capable of training the model to distinguish dozens of object classes. The model's scope was broadened, from simple navigation tasks to a comprehensive analysis of the surrounding environment.

Stage 4. Camera contamination analysis

For the model to work correctly, it was essential to account for the data quality from the cameras mounted on the tractors. During operation, the lenses often became contaminated with dust, dirt, or moisture, which could distort the images and reduce object recognition accuracy. The next task, therefore, was to determine whether the data from a given camera could be used for operational autonomous driving, or if the lens required cleaning.

Keymakr helped Bluewhite prepare an extensive dataset, enabling the model to automatically distinguish between clean and contaminated frames. This solution ensured the filtering of low-quality data, improved system stability, and allowed the client to maintain reliable model performance under real-world operating conditions.

Results

The collaboration between Bluewhite and Keymakr became a clear example of how innovative approaches and team flexibility can transform a complex yet highly meaningful project.
During the first year, several parallel subprojects were implemented, each increasing in complexity and together forming a comprehensive solution for autonomous tractor navigation.
The key challenge was working with 3D data under strict constraints. Keymakr proposed an original solution: a custom tool that converted LiDAR’s 3D information into a 2D format suitable for efficient processing. This unconventional approach increased annotation efficiency by 1.75–2x, enabling faster delivery of autonomy solutions to growers.
By partnering with Keymakr, Bluewhite accelerated the development of its advanced perception systems while upholding the highest standards of accuracy and reliability. This collaboration allowed Bluewhite to deliver autonomy solutions to growers faster, empowering them to operate more safely and achieve greater productivity in the field.
“Keymakr has been throughout our collaboration a reliable, cooperative, and professional vendor. The level of service provided by Keymakr, together with flexibility and the ability to adapt to new domains and tasks, enabled the efficient and fast-growing AI stack currently held by Bluewhite, already deployed to operational tractors, enhancing their perception, navigation, and safety capabilities.”

Nati Kligler, Bluewhite AI Team Lead

“Bluewhite is building something real - it’s amazing to see how the datasets we worked on just yesterday are now literally used in the field, in more ways than one! I’m very hopeful that our partnership with Bluewhite will continue to grow and scale, and that we’ll remain a part of the technologies shaping the future of autonomous agricultural machinery.”

Gleb Zakharov, Keymakr PM

Reviews
on

down-line
high perfomer
high perfomer emea
leader
star
star
star
star
star

"Delivering Quality and Excellence"

The upside of working with Keymakr is their strategy to annotations. You are given a sample of work to correct before they begin on the big batches. This saves all parties time and...

Read more
star
star
star
star
star

"Great service, fair price"

Ability to accommodate different and not consistent workflows.
Ability to scale up as well as scale down.
All the data was in the custom format that...

Read more
star
star
star
star
star

"Awesome Labeling for ML"

I have worked with Keymakr for about 2 years on several segmentation tasks.
They always provide excellent edge alignment, consistency, and speed...

Read more

More
cases

down-line

Automotive

Delivering scalable traffic detection data for a major automotive company

Read more

MusicTech

Providing consensus-based media monitoring to help SoundAware protect the intellectual property rights of creators

Read more

Robotics

Helping Cognex with high-quality OCR and object detection datasets for industrial automation

Read more