Everything a mobile robot does depends on knowing where it is. We design localisation around your accuracy target, sensors and operating environment - from precise positioning in warehouse racks to navigation across large, changing sites.
Localisation connects the robot's map to the job in front of it. Making that estimate useful across a whole site takes more than a successful mapping demo. Three things need to hold together:
Accuracy needs to match the task. Travel through an aisle, alignment at a dock and positioning inside a rack have different tolerances. Those targets need to be met at the robot's operating speed and within its compute budget.
The estimate needs to survive imperfect measurements. Wheels slip, cameras lose features and other sensors drift or become less informative. Sensor fusion, initialisation and recovery need to account for those limits, particularly in repetitive environments.
Maps need to keep up with the site. Pallets move, racks change and new sites open. Mapping, calibration and controlled updates need to be practical for the people installing and operating the fleet.
That is an opportunity to make precise positioning and dependable navigation part of everyday operation, with a localisation system the team can verify and maintain.
02 / HOW WE SOLVE IT
Designed for your environment, your sensors and your accuracy target
We have built localisation for cleaning robots, high-speed warehouse robots, forklifts, supermarket fleets, drones and handheld devices - with our own LiDAR and VSLAM stacks, with purpose-built estimators, and with open-source and vendor stacks (Cartographer, ORB-SLAM3, RTAB-Map, NVIDIA cuVSLAM):
Requirements and sensitivity analysis first. Accuracy per zone, availability, worst-case scenarios - then a simulation-based study of how sensor noise, calibration errors and structural deformation propagate to the position estimate, so the design targets are grounded before hardware exists.
The right sensor set. 2D/3D LiDAR, mono/stereo/fisheye cameras, fiducial markers where the site allows them, IMU, wheel odometry, GNSS/RTK - chosen for the environment and the cost target.
Estimators built for the robot. Extended Kalman filters and factor-graph optimisation with low-latency state prediction for the controller; explicit handling of encoder slip, sensor noise and latency; online extrinsic calibration where drift is unavoidable.
Autonomous mapping and lifelong maps. Robots that map a new site themselves, controlled map updates and change detection, and tooling for operators - to reduce the manual effort of mapping and maintaining sites at scale.
Verified, embedded, delivered. Simulation with ground truth for automated tests, replay of recorded data, validation against ground truth on site; deployment on your embedded platform. Delivered as fully tested, industrial-grade source code - with all rights included when you want to own it.
The goal is a position estimate that meets the task's accuracy and availability targets, with mapping, calibration and tests your team can run throughout the robot's life.
03 / IN PRACTICE
In practice: precise localisation across a 100,000 m² warehouse
Situation. A French AMR builder's fast-moving warehouse robots operated in large, highly repetitive warehouses - up to 100,000 m² - with only millimetres of lateral clearance. The localisation system needed to be designed for that environment and accuracy target.
What we did. We designed the 2D localisation software from scratch: multiple cameras observing fiducial markers, an Extended Kalman filter fusing camera observations with odometry, and a simulation and replay toolchain for automated validation.
Outcome. The customer received a fully tested, industrial-grade implementation - source code and all rights included - delivered through their own repository and review process.
Also: An autonomous mapping procedure lets the robot build the map of a new warehouse itself.
04 / WHAT YOU GET
What you get
A requirements and sensitivity study with a development plan to reach your accuracy target, including risks.
A localisation stack designed for your robot and site (ROS 2 or your own framework), tuned and verified against ground truth.
Mapping and calibration tooling for installation, updates and monitoring.
Industrial-grade source code, tests and documentation - yours to own if you wish.
In this blog post, we'll be taking a deep dive into how to map large indoor spaces using Google Cartographer. We'll cover why you might want to use Cartographer and how to tune it for your specific needs. By the end of this post, you should have a basic understanding of what Cartographer needs and how to get started using it in your own projects.
