TL;DR:
A warehouse robot combines a map, a way to know where it is, a path planner and rules for sharing aisles.
Floor markers or other landmarks fix its position, and a planner finds the shortest route through free aisles.
Safety zones are sized from stopping distance, which depends on speed, reaction time and braking.
A warehouse is a good place to start learning about robot navigation because it is organized. Aisles are straight, shelves stay put, and the floor is flat. Even so, a robot has to answer three questions all the time: where am I, where am I going, and what is in the way?
Maps, markers and traffic rules
The map is usually a grid or a graph of the free floor, with shelves marked as blocked. Many warehouse systems keep it simple on purpose: aisles are lanes, and intersections are nodes.
To know where it is, a robot needs a fix on its position. Wheel odometry alone drifts, so robots correct it with landmarks: markers on the floor or ceiling that a camera can read, magnetic tape or lines they follow, or matching their LiDAR scan against the stored map.
Traffic rules keep several robots from blocking each other. Common ones are one-way aisles, right-of-way at intersections, and reserving a cell or aisle section before entering it, so two robots never plan the same space at the same time.
Worked example: a robot finding its way between shelves
Model the floor as a grid 7 cells wide and 5 cells tall, with columns numbered 0 to 6 and rows 0 to 4, each cell 1 m. Two shelf blocks occupy column 2 and column 4 in rows 1, 2 and 3. Rows 0 and 4 are open cross-aisles, and columns 0, 1, 3, 5 and 6 are open aisles. The robot starts at (0, 2) and must reach (6, 2). Moves are up, down, left or right.
The straight line along row 2 is blocked by the shelves in columns 2 and 4.
To pass column 2, the robot must be in row 0 or row 4. Going up to row 0 takes
2moves.Along row 0 from column 0 to column 6 takes
6moves.Down from row 0 to row 2 at column 6 takes
2moves.Total:
2 + 6 + 2 = 10moves, or 10 m.
Is 10 the shortest? Any route must visit row 0 or row 4, which costs at least 2 vertical moves to get there and 2 to come back, plus the 6 horizontal moves needed to go from column 0 to column 6: 4 + 6 = 10. So 10 is optimal, and the route through row 4 is equally short. A planner such as breadth-first search or A* finds this automatically by exploring free cells in order of distance.
A traffic rule can now settle the tie. Say row 0 is one-way eastbound and row 4 is one-way westbound. A robot going east takes row 0, and one coming back takes row 4. Two robots never meet head-on in a cross-aisle. At 1.0 m/s, the trip takes 10 s plus time for turns and slowing down.
Safety zones from stopping distance
Robots that share floors with people slow or stop when something enters a zone around them. The zone must be bigger than the distance the robot needs to stop. The stopping distance has two parts: distance travelled during the reaction time, v × t_r, and distance during braking, v² / (2a).
For a robot moving at 1.5 m/s with a reaction time of 0.2 s and a braking deceleration of 1.0 m/s², the reaction part is 1.5 × 0.2 = 0.30 m, and the braking part is 1.5² / (2 × 1.0) = 2.25 / 2 = 1.125 m. The total is 0.30 + 1.125 = 1.425 m, so a stop zone of at least 1.5 m is needed, more if the load is heavy or the floor is slippery. Slowing the robot to 0.5 m/s near people cuts the braking part to 0.25 / 2 = 0.125 m, and the reaction part to 0.10 m.
Because distance grows with the square of speed, halving the speed more than halves the stopping distance. That is why slow zones near people are so effective.
Working near mobile robots: Never rely on a robot's sensors alone to avoid you. Follow the site's marked paths, keep clear of the safety zones, and know where the emergency stop is. A loaded robot carries more momentum and takes longer to stop.
Where it goes wrong
Lost position. Markers that are dirty, worn or blocked leave odometry to drift.
Blocked aisles. A pallet left in an aisle means the map is wrong. The planner must re-route or ask for help.
Deadlock. Two robots each waiting for the other to move. Reservation rules and timeouts prevent it.
Wheel slip. Spilled liquid or dust lowers friction, which lengthens stopping distance.
Sensor blind spots. A low pallet corner can sit below a LiDAR's scan height.
How real robots use it
The planner is split into layers. A global planner finds a route like the one above. A local planner steers around people and unexpected objects for the next few meters. A supervisor assigns jobs and manages traffic across the whole fleet. These same ideas show up in robot vacuums and, at higher speeds, in self-driving cars.
Try it yourself
Draw the 7 by 5 grid on paper. Number each free cell with its distance from the start, spreading outward one step at a time: this is breadth-first search done by hand. When the number reaches the goal, walk backward through decreasing numbers to read off the path. Then block one more cell and redo it.
Frequently asked questions
Why not let each robot plan on its own?
They can, but without shared rules they collide or block each other. A central or shared reservation system keeps routes from overlapping.
Do all warehouse robots use LiDAR?
No. Some follow tape or read markers with a camera. The choice depends on how flexible the layout must be.
Get started
