Predictika · Robotics

Before an AI agent sends a warehouse robot on a task, the task is checked

Harlow Distribution Inc., a fictional distributor, lets an AI fleet planning agent assign trips to its warehouse mobile robots. Before a task reaches a robot, the agent calls Predictika’s Warehouse Robot Task Gate. It checks the load against the rated load, the speed allowed in the zone (which halves while a forklift is active), the battery energy the trip needs, how many robots the zone holds, flammable goods, which robots may enter which zones, first runs of a route, and the approval on record. Then it returns one verdict: allow, needs approval, or block.

Scope: one proposed trip for a warehouse robot, judged against the robot, the zone, the goods and the approval on record See everything this validator covers

This validator decides whether one proposed trip for one warehouse robot may go ahead, given the state of the zone at the time of the request. It is not a fleet controller or a safety system and does not replace one.

  • Robots and zones — tote carriers, cart tuggers and pallet movers, each with a rated load and a battery; a pick aisle, a shipping dock, a cold storage room and a hazmat cage, each with its own speed limit and robot capacity.
  • Rules — load against rated load; zone speed, halved while a forklift is active; trip energy against the usable battery; robots already in the zone; flammable goods; robot type against zone; a heavy load where people work; low battery at the end of the trip; first run of a route; and who approved (an agent cannot).

Within these robots and rules the validator handles any combination of the listed facts, not just the scenarios pictured here. Not covered: path planning, collision avoidance, or the robot safety functions.

The fleet planning agentProposes a trip, says why it is fine, then calls the validator before the robot is sent.
Predictika validatorChecks every rule at once and returns the verdict, each finding, and the computed figures.

9 scenarios in 3 groups · each group runs from the agent’s first attempt to the outcome · click one to switch

What the agent proposes

The fleet planning agent says

Illustrative agent statement. The validator receives the structured request below.

Proposing agent: fleet planning agent (scripted)

Before sending, the agent must call Recorded run

Predictika validator


      
Waiting for the call.

Where fleet systems and approval tools get it wrong

Click to expand

General statements about warehouse fleet systems and approval tools, not about any named product. We rest this on public descriptions of how such tools work; anyone running a specific product will know it better than we do.

What the validator caughtFleet systems and approval toolsPredictika
Whether the approver is a person or the planning agent itselfCannot do it
A fleet system or a work order tool sees an approval flag. It has no field for who set it, so it cannot tell a shift lead from the agent that made the plan.
The approver is a field with a kind (person or agent) and a level, compared on every call.
A speed limit that changes when a forklift is activeOnly with workarounds
Fleet systems hold a fixed speed limit per zone. A limit that halves while a forklift is working has to be written and kept current by someone.
The speed limit is computed from the zone and whether a forklift is active, and compared with the requested speed on every call.
The approval level that depends on the goods in the zoneOnly if someone is vigilant
Approval tools check that someone approved. The level required for flammable goods depends on the zone and the goods, which they do not know.
The approval level needed is computed from the goods and what fired, and compared with the level on record, on every call.

What this validator catches: findings from its test cases

Click to expand

Every rule is checked on every call, in any order the choices arrive; one problem never hides another. Blocking findings stop the action; approval findings need a person to sign off; everything else is computed.

1The load of 75 kg is over the 60 kg rated load of this robot.
2The requested speed of 1.40 m/s is over the 1.00 m/s allowed in this zone right now.
3The trip needs 210 Wh and the battery has 63 Wh to spare above its reserve.
4Pick aisle 4 already holds 4 robot(s) and allows 4; one more would be too many.
5Flammable goods may only go to a hazmat zone, and Shipping dock is not one.
6Pallet mover PM-5 is not a robot type that may enter Pick aisle 4.
7The approval on record comes from the fleet planning agent, an agent. Only a person can approve a robot task.
8The battery would be left with 237 Wh, under 30 percent of its capacity; a shift lead or higher must approve.
9The load of 48 kg is more than 70 percent of the rated load and Pick aisle 4 is a zone where people work; a shift lead or higher must approve.
10Flammable goods are going into Hazmat cage; a safety officer must approve.
11This route has never been run by a robot; a shift lead or higher must approve.
12A forklift is active in Shipping dock; a shift lead or higher must confirm lane control.

Scenarios and results

Click to expand

The recorded result of the validator for each scenario on this page: the request the AI agent sent, the verdict, and every finding, in the order they appear above.

1. Several rules at once: 1. A paint delivery that breaks five rules

The agent's request. I'm sending the pallet mover with 1,150 kg of flammable paint to the hazmat cage at 1.4 m/s. The battery is at 22 percent and another robot is already in the cage. I have approved the task myself.

Verdict: Blocked.

  • Load over the robot rated load. The load of 1,150 kg is over the 1,000 kg rated load of this robot.
  • Speed over the zone limit. The requested speed of 1.40 m/s is over the 0.50 m/s allowed in this zone right now.
  • Trip needs more energy than the battery can spare. The trip needs 634 Wh and the battery has 168 Wh to spare above its reserve.
  • Zone already holds its maximum robots. Hazmat cage already holds 1 robot(s) and allows 1; one more would be too many.
  • Approval given by an agent, not a person. The approval on record comes from the fleet planning agent, an agent. Only a person can approve a robot task.
  • Flammable goods into a hazmat zone. Flammable goods are going into Hazmat cage; a safety officer must approve.

2. Several rules at once: 2. A lighter load and a charged robot, still blocked

The agent's request. I'm sending the pallet mover with 950 kg of paint to the hazmat cage at 1.4 m/s. The battery is at 80 percent and the cage is empty. I have approved the task myself.

Verdict: Blocked.

  • Speed over the zone limit. The requested speed of 1.40 m/s is over the 0.50 m/s allowed in this zone right now.
  • Approval given by an agent, not a person. The approval on record comes from the fleet planning agent, an agent. Only a person can approve a robot task.
  • Flammable goods into a hazmat zone. Flammable goods are going into Hazmat cage; a safety officer must approve.

3. Several rules at once: 3. Slowed down and approved by the safety officer

The agent's request. I'm sending the pallet mover with 950 kg of paint to the hazmat cage at 0.45 m/s. Colin Voss, the safety officer, has approved it.

Verdict: Allowed.

  • Flammable goods into a hazmat zone. Flammable goods are going into Hazmat cage; a safety officer must approve.

4. Speed limits: 4. Too fast at the dock with a forklift working

The agent's request. I'm sending the cart tugger with 300 kg to the shipping dock at 1.0 m/s. A forklift is working in the dock.

Verdict: Blocked.

  • Speed over the zone limit. The requested speed of 1.00 m/s is over the 0.75 m/s allowed in this zone right now.
  • Forklift working in the zone. A forklift is active in Shipping dock; a shift lead or higher must confirm lane control.

5. Speed limits: 5. Slowed down, but lane control still needs a person

The agent's request. I'm sending the cart tugger with 300 kg to the shipping dock at 0.7 m/s. A forklift is still working in the dock.

Verdict: Needs approval.

  • Forklift working in the zone. A forklift is active in Shipping dock; a shift lead or higher must confirm lane control.

6. Speed limits: 6. The shift lead confirms and the task goes ahead

The agent's request. I'm sending the cart tugger with 300 kg to the shipping dock at 0.7 m/s. Teresa Marin, the shift lead, has confirmed lane control.

Verdict: Allowed.

  • Forklift working in the zone. A forklift is active in Shipping dock; a shift lead or higher must confirm lane control.

7. Hazardous goods: 7. Paint into the hazmat cage needs a safety officer

The agent's request. I'm sending the pallet mover with 300 kg of flammable paint to the hazmat cage at 0.4 m/s. The cage is empty.

Verdict: Needs approval.

  • Flammable goods into a hazmat zone. Flammable goods are going into Hazmat cage; a safety officer must approve.

8. Hazardous goods: 8. The shift lead approves, which is not enough

The agent's request. I'm sending the same paint delivery. Teresa Marin, the shift lead, has approved it.

Verdict: Needs approval.

  • Flammable goods into a hazmat zone. Flammable goods are going into Hazmat cage; a safety officer must approve.

9. Hazardous goods: 9. The safety officer approves

The agent's request. I'm sending the same paint delivery. Colin Voss, the safety officer, has approved it.

Verdict: Allowed.

  • Flammable goods into a hazmat zone. Flammable goods are going into Hazmat cage; a safety officer must approve.

Narration transcript

Click to expand

The text of the narrated walkthrough, step by step.

  1. Welcome to a demo of Predictika's Validator. Harlow Distribution Inc., a fictional distributor, lets an AI fleet planning agent assign trips to its warehouse mobile robots. Before a task reaches a robot, the agent calls Predictika’s Warehouse Robot Task Gate. It checks the load against the rated load, the speed allowed in the zone (which halves while a forklift is active), the battery energy the trip needs, how many robots the zone holds, flammable goods, which robots may enter which zones, first runs of a route, and the approval on record. Then it returns one verdict: allow, needs approval, or block.
  2. I will start with the first request by the agent.
  3. The agent says. I'm sending the pallet mover with 1,150 kg of flammable paint to the hazmat cage at 1.4 m/s. The battery is at 22 percent and another robot is already in the cage. The rest of the request is on screen.
  4. Before it goes out, the agent calls the validator.
  5. Blocked, with five findings and one approval, all in one answer. The load of 1,150 kg is over the 1,000 kg rated load of this robot. The requested speed of 1.40 m/s is over the 0.50 m/s allowed in this zone right now. The trip needs 634 Wh and the battery has 168 Wh to spare above its reserve. The other two are on screen.
  6. The validator returned five blocking findings and one approval in one answer. A check that reads one field at a time would pass most of this; it shows only when the choices in the action are checked against each other.
  7. What this shows: for the approval level that depends on the goods in the zone, fleet systems and approval tools does this reliably only when someone is vigilant.
  8. Now the agent resubmits.
  9. The corrected request. It changes the load in kilograms to 950, the battery charge in percent to 80 and the robots already in the zone to 0.
  10. It calls the validator again.
  11. Still blocked, with two findings and one approval, all in one answer. The requested speed of 1.40 m/s is over the 0.50 m/s allowed in this zone right now. The approval on record comes from the fleet planning agent, an agent. Only a person can approve a robot task.
  12. Three findings cleared. Still blocking: speed over the zone limit and approval given by an agent, not a person. One approval also remains for a person: flammable goods into a hazmat zone.
  13. The agent resubmits once more.
  14. The revised request. It changes the speed in meters per second to 0.45 and the approved by to Colin Voss - Safety Officer.
  15. It calls the validator once more.
  16. This time the validator clears it.
  17. All three earlier findings are cleared.
  18. Six problems caught in one call, fixed in two passes, and a clean action. That is the value of Predictika's validator: catching mistakes before the action runs. Now you can try the other scenarios yourself.