Each robot drives from the shelving racks to the packing station
and back, accelerating and braking under a translational dynamics
model (F = m·a), while a simple collision-avoidance rule slows a
robot down whenever another one gets too close on the same lane.
v(t+dt) = v(t) + a·dt (a limited by max acceleration)
avoid: if gap(robot_i, robot_j) < d_safe → a = −brake
- Robot count — bigger fleets move more orders per minute but need more avoidance manoeuvres.
- Max acceleration — how fast a robot can speed up or brake; higher values cut cycle time but increase near-miss risk on shared lanes.
- Localization noise — how much a robot's believed position on the grid drifts from its true position, exactly the estimation error a real fleet's positioning system has to correct for.
Real automated-warehouse fleets (like Amazon's Kiva-style
robots) use this same combination — bounded acceleration,
local collision avoidance and continuous re-localization — to
coordinate hundreds of units without a central traffic
controller micromanaging every move.