Task items ride the main conveyor to a single robotic-arm station, which is the physical stand-in for a software bot running a fixed workflow. The arm grabs each item, holds it for one processing cycle, and routes it: on success the item continues down the belt to the exit chute; on an exception it either gets swung onto the return lane — which physically carries it back upstream to re-enter the main belt for another pass — or, once retries are exhausted, dropped straight into the reject bin.
Workflow = Sequence of Automated Tasks + Trigger Conditions
per arm cycle: P(exception) = exception rate
exception & retries < limit → return lane → re-enters belt, retries++
exception & retries ≥ limit → reject bin
no exception → exit chute (completed)
throughput = items exiting / elapsed time (items/min)
success rate = completed / (completed + rejected)
- Item arrival rate — how often a new task item spawns at the belt's start.
- Arm cycle speed — how fast the single arm processes whatever item it is holding; the arm is the bottleneck, same as one bot in a real RPA queue.
- Exception rate — chance a held item needs a retry instead of passing straight through.
- Retry limit — how many times the return lane will carry an item back to the arm before it is rejected outright.