Distributed vs. centralized system in AMR robot fleet management

AMRs (Autonomous Mobile Robots) are quickly becoming a staple of many industries, especially those focused on manufacturing and storage of goods. For the past few years, AMR fleet management was solely based on centralized computing models. This approach is great for highly structured, pre-defined environments, but can fall short in dynamic, less predictable facilities. For such circumstances, distributed computing will do better. But is it this simple? What are the strong and weak points of these models? Are these two management options mutually exclusive? Let’s find out below.

Requirements for AMR fleets

Modern AMR fleets are no longer here for just efficiency – they are also expected to be agile and flexible, which enables them to perform various tasks and quickly adapt to changing working conditions. A good manufacturer of autonomous mobile robots knows this and creates machines that can meet the demands of the modern market.

Each robot should be able to:

• Be cost-effective when compared to human labour;
• Pick optimal paths that allow for avoiding congestions in dynamically changing work environments;
• Effectively carry out their intralogistics tasks, such as picking, moving around, and dropping cargo;
• Adjust their charging breaks in an opportunistic, optimal way to meet the peak requirements each day.

Making all of these possible requires sufficient processing power that allows for instant decision-making on two levels – for the whole fleet and each individual robot.

Distributed and centralized computing systems – what are these?

In distributed computing, each AMR has access to all available data, updates the information network with the new data, and makes its own decisions, based on independent analysis of said data.

As the name suggests, centralized computing utilizes a singular central computer (on-location or cloud-based) that does all the decision-making, performs every analysis and sends out tasks to individual robots. If an issue arises (e.g. a blocked path or loading station), affected AMRs will cease their current activity and await for new instructions.

Pros and cons of distributed software model

Pros:

• No need for a large, resources-consuming central data centre;
• Better resource optimization thanks to distributed memory computers and appropriate distribution strategies;
• On-the-spot decisions and almost instant path planning for autonomous robot fleets;
• Distributed memory machines are less impacted by connectivity issues and can operate on their own with just basic instructions;
• Distributed autonomous systems are easily scalable and can be more robust, as there’s no single point of failure that can paralyse the entire network.

Cons:

• Due to being complex, structured adaptive mesh methods require knowledge of distributed databases and parallel processing;
• The lack of a central authority can lead to issues when conflicts appear (e.g. when two robots decide to do the same task).

Pros and cons of centralized software model

Pros:

• The presence of a singular master authority helps in resolving conflicts and fleet behaviour optimization, which theoretically leads to optimal results;
• Centralized computing can be done completely off-site thanks to cloud options;
• A central computer can synchronize between different data sources and robot types, allowing the use of more varied operational tools.

Cons:

• High cost of maintaining the necessary processing power and connectivity for all involved devices;
• Robots require constant communication with the central computer, which can put heavy strain on the facility’s bandwidth;
• Possible gaps in network connectivity can lead to robots getting “stuck” in “dead zones” and requiring on-site human intervention.

What’s the best option?

As you can see, both options have their strong and weak points. Which is better, then? Well, the answer isn’t as black and white as you might expect. According to the robotics community, the best solution would be to combine the two methods into a hybrid one. With this, you can largely mitigate the shortcomings of both computing models without significantly raising costs.

For example, the centralized component can be used as a digital supervisor of sorts – providing general directions while leaving a high level of autonomy for the distributed network of robots. This would help in resolving conflicts while eliminating the need for full connectivity coverage across the whole facility – robots would perform pre-set, centrally issued tasks while using their own computing power to make real-time decisions when the need arises.

Automation integrators and AMR algorithm providers know this and prepare digital platforms that allow such a hybrid approach and require minimal human supervision. Thanks to this, full industrial automation is becoming more available almost daily. Technologies such as Digital Twin allow for e.g. parallel adaptive simulation of various solutions before implementation in the real facility or set an easily reproducible standard benchmark for each robot’s performance.

AMR fleet management – a summary

Modern AMR fleet management is often based on shared memory architectures (distributed computing methods) as extensive networks of robots would require massive resources to maintain an entire AMR hierarchy governed from a single point (centralized computing). Using a central computer in such cases would put an artificial handicap on the robot fleet. However, some parts of centralized governing (such as resolution of conflicts or assigning specific robots to do specific tasks) can be beneficial to any fleet. With this in mind, the implemented computing method should be a mixture of both options – the best solution would be a robot fleet with high individual autonomy but with some douse of central governing. Thus, robots would be free to make on-the-spot decisions while performing their pre-set objectives without the risk of more than one unit deciding to do the same thing simultaneously. How to achieve this? It’s simple – rely on trustworthy, experienced software designers and automation integrators. With this, you can be sure that the fleet management platform you get will fit your company’s needs like a glove. Plus, thanks to modular structure, such software can be easily modified later on if you decide to expand your AMR fleet.