Robots and People: Keeping Them Apart, or Letting Them Share

Munro Footwear Group in Melbourne runs 205 Mushiny robots across 5,239 shelf pods, and no picker walks among them. The robots work inside a fenced area. The people stand at stations around the outside, and the stock is brought to them. That is how a shelf-to-person system is built, and it is a safety decision as much as a layout decision.

There are two ways to put robots in a warehouse. You can keep people and robots apart, or you can let them share the same space. The choice looks like a safety question. It is also a navigation question, a speed question and a cost question, and it is the same question each time. This article explains the two designs, what each one buys you, what each one costs, and what to ask a supplier before you decide.

What are the two designs?

Keep them apart. A goods-to-person grid is fenced. The robots have the floor to themselves, and people work at stations on the outside. Nobody needs to cross the grid to do their job, because the stock comes to them.

Let them share. Some work cannot be fenced. Where people walk through, pallets get left in aisles and stock is staged somewhere new, the space changes. Here the robots and the people use the same floor.

Mushiny builds for both. The grid robots in shelf-to-person, MIX 4.0 and MIX Pro work in the fenced design. The M6-300 is Mushiny's SLAM variant, built to work in a space it shares with people.

Why does the design decide how the robot navigates?

Because the robot can only be as simple as the space it works in.

A robot in a fenced grid navigates by QR code. The floor carries a grid of QR markers. The robot reads them as it travels and knows exactly where it is, to the millimetre. The obstacles it plans around are a known, fixed set: the racking, the pods and the other robots. Nothing unexpected walks in.

A robot that shares space with people cannot rely on that. It uses SLAM, which stands for simultaneous localisation and mapping. The robot scans its surroundings, works out where it is from what it sees, and updates its map as it goes. It needs to, because a person, a pallet or a stack of cartons can be somewhere today that they were not yesterday.

So the safety design and the navigation design are one decision. If you can keep people out, you can use the simpler, faster, more accurate way of finding your position. If you cannot, you need the robot to build and keep checking its own picture of the room.

Is a QR marker a track?

No.

A QR marker tells the robot where it is. It does not tell the robot where to go. A fixed-path vehicle follows a route that was set at installation, and when something blocks the route it stops and waits. A grid robot is given a destination and works out its own route across the grid on the day. It negotiates with the other robots around it and routes around them.

Why does isolation make robots faster and more accurate, not slower?

When the robot knows its position exactly, it can drive at speed without stopping to check. When it has to keep re-checking what it can see, it runs slower. The marker on the floor also does not drift. A map built from scanning can be good, but it depends on what the robot can see at that moment.

Cost follows the same logic. A QR grid robot costs less per robot than a SLAM robot.

So the fence is not what slows the system down. The fence is what lets you use the faster method.

How do the two designs compare?

What does the operator's day look like in each?

In a fenced grid, the operator works at a station. The picker does not walk to the stock. The robot lifts a pod, carries it to the station and presents it. In MIX 4.0 a workstation presents up to 240 totes an hour, holds 36 buffer locations for hot items, and runs two queuing positions and two picking positions, so the operator picks continuously rather than waiting for the next pod. The job is picking. Nothing in it involves watching for a robot.

In a shared space, people and robots work in the same area. The robot is built to cope with that: it keeps scanning and routes around what it finds, including people. The trade is the speed, cost and precision set out in the table.

In the grid, the robot never picks. A person does, at a station, and the robot's job is to keep that person busy.

Is keeping people out a constraint as well as a benefit?

Yes.

The fenced grid works because the floor belongs to the robots. That means you give up the floor. People are kept out of the area, so it cannot double as a route for anyone on foot. You plan the building around the grid, with the stations on the outside.

It also means the grid is the layout. If you move the racking, the grid is re-laid out and remapped. A SLAM robot updates its map instead. If your floor changes often, or the work cannot be moved to the edge of a fenced area, the fenced design is the wrong answer and you should say so early.

What happens when someone has to go into the grid?

Sometimes someone will. A robot stops, a pod needs attention, or a fault needs a person on the floor.

The T6 series carries LIDAR, a bump stop, an e-stop and zoned access. Those are the features on the robot and in the system. How entry is controlled on your site is a procedure we work out with you and demonstrate before go-live, because it depends on your layout and your own safety rules.

One design point does help. In a grid, if one robot stops, the others keep working. The system has no conveyor and no single point of failure, so a stopped robot does not stop the site. That gives you time to deal with it properly rather than in a hurry.

Ask to be walked through this case, not just the normal running case. See the questions below.

What should you ask a supplier?

Take these to any supplier, including us.

1. How does the robot know where it is? Markers on the floor, or its own map, or both. The answer decides whether the robot can run fast and whether it can share space.
2. Is the robot designed to share space with people, or designed to be kept away from them? The two answers lead to different systems. Do not accept a vague answer.
3. What does the system do when a robot meets something it did not expect? In a fenced grid the answer should be that this is rare by design. In a shared space it should be a normal case the robot handles.
4. What happens when someone has to enter the robot area? Ask to see it done, step by step, including how the robots in that area are stopped and how they restart.
5. What happens when one robot stops? In a grid built as a swarm, the others should keep working.
6. What are the safety features on the robot, and where are they documented? Ask for the specification in writing and keep it with your own risk assessment.

You can see how Mushiny's grid behaves at the Chatswood Tech Centre, the only place in Australia with the full range running live under one roof. It is free to visit, and you can enquire here. The references page lists the live sites.

How do you choose between the two designs?

Start with your floor, not the technology.

If the work is picking and storage, and you can give the robots an area of their own, the fenced design is the better answer. You get speed, accuracy and a lower cost per robot, and the people work at stations. Shelf-to-person, MIX 4.0 and MIX Pro are all built this way.

If the robots have to move through areas people use, and you cannot fence them off, you need a robot that is built to share. Expect to pay for that in speed, cost and precision. It is the right trade when the alternative is rebuilding the way the building works.

Frequently asked questions