Useful Information About Warehouse Robots

Hello! I’m Robert, the founder of YaCu Robotics. We develop a “Driver” program that can automate any wheeled vehicle. For example, we have automated a Toyota Prius and are testing it in closed environments. We have also designed and built an autonomous floor-cleaning machine for commercial spaces. We collaborate with robot manufacturers and integrate our “Driver” into their systems. At the same time, we act as a system integrator of warehouse robots from third-party manufacturers. Our deep expertise in autonomous technologies allows us to deliver end-to-end warehouse automation solutions.

I would like to share my knowledge in this field from a technical perspective, as well as our experience of more than 5 years. Therefore, everything stated below is based исключительно on personal experience.

A brief introduction. Five years ago, when autonomous technologies were not as widely discussed, robots and their variations were mostly a hype topic. I strongly believed that the future would arrive faster than expected and that self-driving cars were only a matter of time. After sharing these thoughts with my technical partner, we decided to carefully start exploring this field, with the goal that when large-scale automation arrives, we would already have deep expertise. That’s how we began developing our concept.

To start, I purchased a Toyota Prius and a device capable of extracting raw data from the CAN bus. We needed to retrieve steering angles, acceleration, and braking data. The Prius was chosen as the most affordable vehicle where full control interception had already been achieved, since all its controls are electronic.

After learning how to extract data, we began developing a neural network and introduced a concept of focused attention, similar to human vision. When we drive, we perceive everything through peripheral vision and shift focus only to relevant events. By equipping the car with a stereo camera and connecting a computer, we started collecting datasets for training. Once we achieved full control over the vehicle, the main question of autonomy arose: how to organize navigation—how the machine determines its current position and destination. At that time, we still didn’t know what our final product would be.

There was no single correct navigation standard then, and there still isn’t today. But this is exactly where the most interesting research happens—and where breakthrough technologies emerge.

How SLAM works

From robotics came a technology called SLAM (Simultaneous Localization and Mapping). In simple terms, a LiDAR scans the environment and compares it with stored patterns. If they match, the robot understands where it is. In practice, however, this is much more complex—data must be constantly correlated with wheel odometry to estimate distance traveled and correct positioning errors.

This technology has several drawbacks. For example, poor performance in large areas, high memory requirements, and difficulties in dynamic environments. If a warehouse layout changes significantly (e.g., new pallets), the robot may interpret it as a new space and lose orientation. SLAM also struggles in open environments due to limited LiDAR range. Additionally, route planning decisions are made by the robot rather than the operator.

AGV robots

Due to the complexity of SLAM, the first warehouse robots were AGVs. These are cart-like robots that move strictly along predefined routes—usually physical ones such as magnetic strips, QR codes, or laser guides. They typically cannot avoid obstacles, only stop. This creates infrastructure limitations and reduces flexibility. AGVs are difficult to scale and are mainly viable for large companies with predictable, repetitive workflows. For example, Amazon optimized shelf delivery to pickers using such systems.

This type of automation suits operations similar to Amazon’s. A 6,000–10,000 m² warehouse would require about 100 robots. Each robot costs around 1.3 million rubles. Additional costs include specialized shelving and reduced storage density. Suitable warehouses are low-rise with multiple floors—relatively rare. In Russia, Ronavi produces such robots, comparable to China’s Geek+. YaCu Robotics integrates warehouse robots from Seer, a leading company with over 20 robot types.

Robot navigation

Understanding SLAM limitations, we explored more stable autonomy solutions. In autonomous vehicles, GNSS provides localization with centimeter-level accuracy. However, it only works outdoors. We decided to create an indoor equivalent—a GPS-like system using RTLS beacons. After writing tens of thousands of lines of code, we developed a system where beacons are placed around a facility, and a tag on the robot provides precise location and direction data. This is global navigation.

Without false modesty, this is an extremely reliable system with no direct analogs. With a predefined map, the robot can move freely to any point. This flexibility enabled the emergence of AMR robots.

AMR robots

These are the next generation after AGVs. They use more sensors, primarily LiDAR, which scans surroundings to build maps and avoid obstacles. Unlike automotive-grade LiDARs, warehouse versions are simpler and cheaper. Additional sensors include stereo cameras and ultrasonic sensors.

This complexity supports SLAM-based navigation, allowing robots to operate in dynamic environments. AMRs can function in human-robot mixed environments, though typically slower than AGVs. They come in many formats: shelf movers, transport platforms, or picking assistants. For example, DHL is actively implementing such robots.

Autonomous forklift

Autonomous forklifts use similar technologies but introduce additional complexity. Pallet handling requires high precision, consistent pallet quality, and accurate load placement—challenging in practice. While promising, full automation is still limited.

However, forklifts are ideal for transporting pallets between zones, reducing manual labor. Such robots cost around 5 million rubles. We are currently developing a hybrid solution where humans handle pallet pickup, and robots handle transportation. This simplifies deployment and eliminates the need for WMS integration.

In the next article, I will cover RCS (Robot Control Systems) and how robots receive tasks from WMS.