Emergency Stop

The E-Stop must function as the highest priority interrupt in the system, bypassing all standard logic, queues, and resource locks.

• Non-maskable Interrupt (NMI): The stop signal cannot be ignored or delayed by the agent's cognitive loop.
• Preemption: It immediately suspends all concurrent processes, including tool calls and API executions.
• Example: In ROS 2, a hardware E-Stop triggers a SIGINT chain that preempts the executor's callback queue, halting motor commands within microseconds.

A single point of failure negates the purpose of a safety mechanism. Effective E-Stops employ dual-channel architecture.

• Dual-Channel Logic: Two independent circuits (e.g., Cat 3/4 per ISO 13849) must agree; a mismatch triggers a fail-safe state.
• Diverse Triggers: Combines a physical mushroom button with a software-based Poison Pill Message and a Watchdog Timer timeout.
• Example: An autonomous mobile robot (AMR) uses a safety-rated PLC that monitors both the physical button state and the heartbeat of the main compute module.

The system must default to a safe state upon any failure in the E-Stop mechanism itself, including loss of power.

• De-energize to Trip: The safety circuit is energized to permit motion. A break in the circuit (button press, cut cable, power loss) instantly removes power.
• Fail-Closed Configuration: Actuators default to a braking or locked position rather than free-spinning.
• Example: A robotic arm's joint brakes are held open by solenoid power. An E-Stop event cuts the solenoid circuit, causing the spring-loaded brakes to engage mechanically.

The system must not automatically restart upon release of the E-Stop button. A deliberate, secondary manual action is required.

• Anti-Restart Logic: The safety controller latches the stop signal until a dedicated reset button is pressed.
• State Validation: The reset sequence verifies that the hazardous condition has cleared and all systems are in a Readiness Probe state.
• Example: After an E-Stop, an industrial robot controller displays a 'Safety Stop Active' status and requires the operator to physically rotate and release the mushroom button, then press a blue reset button on the HMI.

The actuator must be instantly recognizable and operable under panic conditions without fine motor skills.

• Standardized Design: A red mushroom-head button on a yellow background, compliant with ISO 13850 and IEC 60947-5-5.
• Accessibility: Positioned within immediate reach of the operator's danger zone, free from obstructions.
• Positive Operation: The button provides tactile feedback (snap-action) confirming the circuit has been mechanically forced open, not just a capacitive touch sensor.

An effective E-Stop triggers an immediate Controlled Shutdown Sequence that includes persisting critical state for forensic analysis.

• Atomic State Capture: An Immutable State Snapshot of the agent's memory and trajectory is saved to non-volatile storage before power is cut.
• Idempotent Rollback: The system logs the exact sequence ID of the last safe state to ensure a clean recovery without replaying dangerous actions.
• Example: A financial trading agent's E-Stop triggers a SIGTERM handler that cancels all open orders via the exchange API and writes the portfolio snapshot to a write-ahead log before exiting.

A mechanism designed to completely and immediately shut down an autonomous agent or system, overriding all other processes to prevent unintended harm. Unlike a graceful degradation, a kill switch prioritizes absolute termination speed over state preservation.

• Can be physical (hardware button) or digital (API command)
• Must bypass all normal shutdown queues and priority checks
• Often implemented as a normally-closed circuit that triggers shutdown when broken

A safety mechanism that automatically triggers a kill command or safe state if the human operator becomes incapacitated or fails to provide a periodic confirmation signal. This is the inverse of an emergency stop—it activates through inaction rather than action.

• Requires continuous heartbeat signals from the operator
• Common in locomotive engineering and industrial robotics
• Prevents runaway operation when human oversight is lost

A hardware or software timer that monitors an agent's operation and triggers a corrective action, such as a system reset, if the agent fails to periodically signal that it is functioning correctly. This addresses silent failures where an agent appears alive but is actually frozen.

• Agent must 'pet' the watchdog at regular intervals
• Failure to signal triggers an automatic system reset or kill
• Implemented in both microcontrollers and cloud orchestration

A software design pattern that prevents an agent from repeatedly attempting an operation that is likely to fail, immediately failing such calls for a set timeout period to allow the system to recover. This prevents cascading failure amplification.

• Three states: Closed (normal), Open (failing fast), Half-Open (testing recovery)
• Protects downstream services from overload
• Differs from an emergency stop by targeting specific operations rather than full termination

A pre-defined, secure condition that an autonomous system automatically enters upon detecting a critical malfunction, designed to minimize potential damage or danger. The emergency stop is the actuator; the fail-safe state is the destination.

• Examples: robotic arm going limp, valve closing, vehicle pulling over
• Must be designed to be inherently stable without active control
• Contrast with fail-secure (prioritizes security) and fail-operational (maintains function)

A control mechanism that requires explicit human approval for an agent's high-stakes actions or allows a human operator to manually veto and halt an agent's ongoing process. This is the decision layer that sits above automated kill switches.

• Can be implemented as approval gates or real-time veto controls
• Must operate on an independent, out-of-band communication channel
• Critical for high-consequence actions like financial transactions or physical actuation