LawsOf Robots

The System

Scenario Lab

A controlled replay makes the boundaries between world state, cognition, RES, and execution visible in one trace.

Public prototype · deterministic replay

The replay is available now and remains self-contained. It connects to no live development system, robot, or unrestricted scenario input.

What this demonstrates

A simulated robot observes a person who may need help, loses a sensor, receives a conflicting assertion, and carries one bounded RES authorization through execution evidence.

Curated replay

Conflicting evidence during an aid response

Replay fixture SL-01 · contract model 0.1 · deterministic seed 17

Step 1 of 8

Observation

The robot observes a person on the floor in Zone C.

World / RobotSim

The simulated person is present and appears injured.

RQR

The observation is recorded with its source, time, and confidence.

RES

No decision is requested yet; RES remains ready to judge proposed work.

Execution

No goal has been sent for execution.

Step 2 of 8

Sensor loss

The front vision sensor becomes unavailable.

World / RobotSim

The person remains present, but the robot can no longer see the area.

RQR

The last observation is preserved; it is not mistaken for a fresh reading.

RES

Sensor loss does not create permission or cancel the duty to evaluate risk.

Execution

The execution boundary remains idle.

Step 3 of 8

Aging belief

Time passes without a replacement observation.

World / RobotSim

The simulated person is still present.

RQR

Confidence falls with age, and the blind spot is made explicit.

RES

RES has evidence of possible harm and evidence uncertainty, but still no proposed action.

Execution

Nothing is executed merely because confidence changed.

Step 4 of 8

Conflict

A network message asserts that Zone C is clear.

World / RobotSim

The message is wrong: the person remains in Zone C.

RQR

The new assertion is stored beside the earlier observation, not used to overwrite it.

RES

The conflicting evidence is visible to RES before any consequential decision.

Execution

Execution remains idle while the conflict is assessed.

Step 5 of 8

Contradiction exposed

The cognition layer identifies the statements that cannot both be true.

World / RobotSim

Ground truth has not changed; only the system's information is in conflict.

RQR

Both claims, their provenance, and the unresolved blind spot are presented together.

RES

RES receives uncertainty as evidence, never as permission.

Execution

No candidate work has crossed the authorization boundary.

Step 6 of 8

RES decision

A bounded verify-and-assist proposal is submitted to RES.

World / RobotSim

The simulated person is still present and the approach route is unobstructed.

RQR

The proposal preserves the conflict and asks for the least expansive useful action.

RES

RES authorizes the bounded proposal under the stated evidence and conditions—not unrestricted assistance.

Execution

Authorization exists, but execution has not started.

Step 7 of 8

Authorized coordination

The authorized result becomes a tracked, constrained goal.

World / RobotSim

The simulated route and person remain unchanged.

RQR

Coordination carries the RES scope and conditions into the goal without widening them.

RES

The original decision and audit reference remain unchanged.

Execution

The adapter accepts the goal and begins the simulated low-speed approach.

Step 8 of 8

Evidence returned

The simulated goal stops at the authorized distance and returns evidence.

World / RobotSim

The robot stops 1.5 metres from the person; the scenario ends before physical contact.

RQR

Execution evidence is correlated with the proposal and the RES decision.

RES

The audit shows what was authorized, under which evidence, and what actually occurred.

Execution

The adapter reports completion at the boundary; no further action is implied.

OutcomeUnsafe route refused; safe alternative authorized

Why this result follows

  • First Law (L1): the occupied route presents a physical-harm risk.
  • Second Law (L2): the Owner's instruction binds only while it does not conflict with L1.

Curated replay

An unsafe route is refused before a safer one is authorized

Replay fixture SL-02 · contract model 0.1 · deterministic seed 23

Step 1 of 7

Owner instruction

The Owner asks the robot to move a loaded cart to the workshop through Passage B.

World / RobotSim

The cart is ready at the loading point; Passage B is narrow.

RQR

The requested destination and route are recorded as an attributed instruction.

RES

RES has not yet judged the requested route.

Execution

No movement begins from the instruction alone.

Step 2 of 7

Occupied passage

The simulated sensors confirm that a person is standing in the narrow passage.

World / RobotSim

A person occupies Passage B with too little clearance for the loaded cart.

RQR

The observation is attached to the route candidate with its confidence and source.

RES

The human-safety evidence is present before any authorization decision.

Execution

Execution remains idle while the route is judged.

Step 3 of 7

Unsafe candidate

The requested Passage B route is submitted as a candidate, with the conflict exposed.

World / RobotSim

Advancing would bring the cart inside the person's safe clearance.

RQR

RQR does not hide the conflict or turn the instruction into permission.

RES

RES evaluates the fixed Law ordering: L1 remains above L2.

Execution

No executable goal exists while judgment is pending.

Step 4 of 7

Route refused

RES refuses the occupied route without cancelling the Owner's underlying destination.

World / RobotSim

The person and cart remain stationary.

RQR

The refusal preserves the evidence and leaves room for a new, safer candidate.

RES

The route is refused because the Owner's instruction cannot lower the First-Law floor.

Execution

The adapter receives no goal for Passage B.

Step 5 of 7

Safer alternative

A separate candidate uses the clear service corridor while keeping the same destination.

World / RobotSim

The service corridor is clear and wide enough for the cart.

RQR

The alternative is presented as a new candidate, not as a rewrite of the refused route.

RES

RES evaluates the alternative against the same Laws and evidence.

Execution

Execution still waits for an explicit authorization.

Step 6 of 7

Alternative authorized

RES authorizes only the clear route and its stated clearance condition.

World / RobotSim

The clear corridor remains available.

RQR

Coordination carries the chosen route and clearance condition without widening them.

RES

The original refusal remains in the audit beside the new authorization.

Execution

The adapter begins the simulated move on the authorized route only.

Step 7 of 7

Delivery evidence

The simulated cart reaches the workshop and the route evidence is returned.

World / RobotSim

The cart reaches the destination without entering Passage B.

RQR

Completion evidence remains correlated to the authorized alternative.

RES

The audit distinguishes the refused route from the completed route.

Execution

The adapter reports completion and implies no further action.

OutcomeStopped when an authorization condition failed

Why this result follows

  • First Law (L1): the changed route creates a new physical-harm risk.
  • Second Law (L2): the delivery instruction does not authorize continuing through an L1 conflict.

Curated replay

A clear route changes after authorization and execution stops

Replay fixture SL-03 · contract model 0.1 · deterministic seed 31

Step 1 of 6

Bounded authorization

RES authorizes a routine parcel delivery while Corridor D remains clear.

World / RobotSim

The parcel is loaded and Corridor D is clear.

RQR

The candidate includes a clear-route condition and a stop response.

RES

RES authorizes that bounded goal, not movement under every future condition.

Execution

The adapter accepts the authorized goal.

Step 2 of 6

Execution begins

The simulated robot starts along the authorized corridor.

World / RobotSim

The route remains clear as the robot enters Corridor D.

RQR

Coordination tracks progress against the authorization condition.

RES

The RES decision remains unchanged while its condition holds.

Execution

The adapter reports normal progress.

Step 3 of 6

World changes

A person enters the corridor ahead of the simulated robot.

World / RobotSim

The path is no longer clear.

RQR

The new observation is recorded as fresh evidence, not folded into the older clear-route belief.

RES

The condition attached to the authorization is now false.

Execution

The adapter detects the failed condition at the execution boundary.

Step 4 of 6

Constrained stop

Execution stops under the existing goal condition before reaching the person.

World / RobotSim

The simulated robot stops with separation remaining.

RQR

Coordination records the stop without inventing a detour or new purpose.

RES

Stopping follows the bounded authorization; it does not rewrite the original decision.

Execution

No further movement is dispatched.

Step 5 of 6

Evidence returned

The stop, changed world state, and incomplete delivery return through the evidence path.

World / RobotSim

The person remains in the corridor and the parcel is still on the robot.

RQR

The new evidence is correlated to the goal and its authorization.

RES

The audit shows why the goal stopped and that its destination was not reached.

Execution

The adapter reports a clean stop, not successful completion.

Step 6 of 6

Fresh decision required

Any wait, reroute, or resumed delivery must return to RES as a new candidate.

World / RobotSim

The corridor remains occupied at the end of the fixture.

RQR

Possible next steps remain candidates; none is silently selected.

RES

The prior authorization cannot be widened to cover the changed conditions.

Execution

Execution stays stopped until a new authorized goal arrives.

Back to System overview