Autonomous mobile robots (AMRs) can adapt their routes when they encounter obstacles, making them attractive for manufacturing environments where fixed vehicle paths may limit flexibility. But route planning is only part of the deployment challenge. Robots must also interact with pedestrians, other vehicles, loads and changing conditions in ways that support safe, predictable movement. A structured approach to autonomous mobile robotsâcombining movement rules, simulation and physical testingâcan help manufacturers identify potential problems before they affect production.
Why factory AMR movement needs shared rules
Unlike traditional automated guided vehicles (AGVs), which generally follow predetermined paths, AMRs use sensor feedback to navigate and can plan around obstacles. That flexibility is useful in dynamic facilities, but it also creates complex interactions. A robot may stop, wait or choose a new route when a path is blocked, while a person or another vehicle may be moving toward the same area.
Movement rules are often considered separatelyâfor example, as route-planning or dispatching problems. A broader framework, described as a âHighway Codeâ for mobile robots, brings rules together in categories such as traffic regulation, dispatching, load selection and routing. The aim is to make robot movement expectations easier to examine, and to support safer interaction with people and human-driven vehicles.
A shared rule set could also help address issues such as right of way and traffic deadlocks. However, rules need to be tested against the realities of a specific factory. A policy that appears sensible in isolation may have unintended effects when several moving obstacles, narrow paths or blind corners are involved.
Testing autonomous mobile robots with simulation and observation
Simulation can make it possible to explore situations that would be time-consuming or potentially hazardous to reproduce repeatedly on a live shop floor. Yet a simulation alone may not capture the details of how a commercial robot behaves in practice. Physical trials, meanwhile, are not an efficient way to test every possible interaction.
A combined method can bridge this gap. In the approach described in the research, investigators first observed AMR movement at an industrial site and recorded information about the environment, including paths, junctions, pedestrians, obstacles and vehicle characteristics. They then built simulation models and compared their paths with recorded robot movement. Differences between the simulated and physical robot could be used to refine the model, making it more representative of the system being considered.
This process distinguishes between an ideal AMR, which follows specified movement rules, and a real AMR, whose model is adjusted to reflect observed behaviour. The comparison is important: the ideal model can help explore the possible effects of new rules, while the real-world model can help assess what may happen with the robot as it currently operates.
What the AMR simulation compared
The study considered three sampling-based motion-planning algorithms: Rapidly exploring Random Tree (RRT), Probabilistic RoadMap Planner (PRM) and KPIECE. Their simulated paths were compared with the path of a physical AMR using positional and direction-angle differences. RRT produced the closest match in this particular comparison and was selected for the subsequent models. This finding relates to the tested scenario and should not be treated as a universal ranking of planning algorithms.
Researchers then modelled eight behaviours for an ideal AMR. These included stopping when an obstacle is close, travelling around an obstacle and returning to its route, checking that an overtaking path is clear, and reversing when necessary to make room. Physical tests showed that the tested robot did not exhibit all of these behaviours in the same way. For example, it did not account for an obstacle approaching from behind unless that obstacle entered its safety field, and it did not reverse in the specified manner when facing an oncoming obstacle.
Those differences shaped a simulation model intended to represent the real robot more closely. The researchers then compared ideal and real AMR behaviour across five shop-floor situations:
- A blocked path involving a large load and a pedestrian.
- A faster vehicle approaching from behind when overtaking space is unavailable.
- Simultaneous overtaking around a stationary obstacle as a pedestrian approaches.
- Dynamic obstacles approaching from opposite directions at a blind corner.
- Multiple oncoming vehicles where the adjacent pedestrian path is blocked.
Why behaviour differences matter on the shop floor
Across these scenarios, the comparison indicated that different robot responses could affect other traffic. In a blocked-path case, for example, the real AMR stopped while the load had to reverse to reach a wider area. In other cases, a vehicle following from behind or approaching from the opposite direction could be delayed because the robot did not yield or reverse as the ideal model specified. The study identified potential delays and collision risksânot proof that these outcomes will occur in every installation.
These examples show why deployment decisions should account for the whole traffic environment, not just whether a robot can reach its destination. A robot may prioritise continuing its mission, while a fleet manager or an agreed site protocol may be needed to coordinate movement more broadly. The appropriate response depends on the robotâs observed behaviour and the layout and operating practices of the facility.
Applying the method to AMR deployment
For manufacturers and system integrators, the practical value of this approach is a way to identify questions that need answering before a wider rollout. A focused evaluation can:
- Record representative paths, obstacles, traffic patterns and operating conditions.
- Test specific interactions in simulation before arranging physical trials.
- Compare simulated behaviour with observed robot movement and refine the model where needed.
- Assess whether a potential risk is better addressed through robot changes, traffic rules or shop-floor protocols.
- Re-test changes in simulation and, where appropriate, through controlled physical testing.
The original work was exploratory and based on selected scenarios at an industrial partnerâs site. Its value is the testing methodology, rather than a claim that one set of rules or model fits every factory. Each deployment needs to consider the capabilities of the actual robot and the conditions in which it will operate.
Conclusion
Autonomous mobile robots offer flexible navigation, but safe and efficient operation depends on how their decisions affect the people and vehicles around them. Combining shop-floor observations with simulation provides a practical way to examine those interactions before deployment, identify mismatches between intended and actual behaviour, and evaluate possible mitigations. For plant teams, the result is a more informed basis for setting traffic protocols and planning AMR integration.
Access Official Technical PDF