Obstacle detection: Harder than you think? 

When it comes to obstacle detection for autonomous mobile robots (AMRs), identifying objects is the easy part.

Segmenting the floor and eliminating false positives are the real challenges today.

In this article, we’ll explore these two concepts in detail. Then, we’ll look at what options are available today.

Why floor segmentation and false positives are problems for AMRs

Floor segmentation and mitigating false positives are important – and challenging – for AMR developers to overcome because:

1. Every facility is different: Your robot must be able to see all floor types and conditions with certainty.
2. False positives create unplanned stops: Human intervention downtime reduces your robot’s availability and performance.

Every facility is different

Different types of floors create a different type of noise signature in the detection system. This is dependent on the physics engine the camera is using and the different floor materials.

Reducing the noise on a nice clean floor is easy. However, that noise must be reduced in any environment so the robot can see even small objects on the ground. Dirt on the floor, or even flooring that is too shiny, can cause a problem.

Similarly, creating a perception platform for a single facility is easy. But, as an OEM you need a versatile, adaptable perception platform that increases your robots' availability and performance.

False positives create unplanned stops

In addition to your robot being able to detect an object, it also needs to be able to identify what that object is and whether or not it needs to stop for it. A robot stopping when it shouldn't isn't as available as one that keeps moving.

When a robot requires human intervention every time it stops, each stop costs your customer money in lost productivity. The impact of that downtime is even more pronounced when a robot stops unexpectedly because of shiny objects, dust or other harmless particles in its path.

Development for adaptable, high-performance solutions

ifm concentrated on floor segmentation and false positive mitigation when we developed our obstacle detection system. Our solution:

• Is adaptable and suitable for multiple environments

• Increases throughput with uninterrupted performance

Adaptable and suitable for multiple environments

An AMR that operates in various environments can be deployed across more industries and applications than its competitors. That flexibility allows organizations to streamline processes, respond to evolving demands, and achieve optimal performance in diverse operational landscapes.

And, the robot can transition between tasks in different areas without the need for reconfiguration – another important feature for end-users. This versatility also reduces the need for multiple robots specialized for specific environments.

Increase throughput with uninterrupted performance

Interruptions in robot operation leads to significant productivity losses and disruptions in workflow. By minimizing false positives, the robot ensures continuous and reliable performance, contributing to consistent efficiency in various tasks.

In dynamic environments, a robot that doesn't halt unnecessarily helps maintain a smooth and uninterrupted workflow, preventing unnecessary delays in operations. This reliability is especially important in industries where precision and timing are critical, such as manufacturing, logistics, and healthcare.

Ultimately, this all adds up to significant value to a prospective customer. You may not be able to bring the price of your AMR down much more. At this point, the price points for pretty much all AMRs are the lowest possible. 

Instead, you can offer better productivity than your competitors. And that means cost-savings for a client. 

Your customer may still pay about the same for each unit. But, they can save hundreds of thousands of dollars overall when you can offer them fewer AMRs that offer the same (or better!) throughput as a larger fleet from a competitor.

<< Why your robots hit stuff | Faster perception stack processing >>