# Intent, Accessibility, and Human Override — descriptive transcript

Complete non-video learning alternative. Original supervisory teaching model, not a safety system. Captions reproduce narration; visual descriptions explain information conveyed by motion. Thresholds are invented lab values.

## 00:00–00:16 · Which way will it go?

**Visible mechanism:** A wheeled cart rolls up to a fork, then holds before choosing a station. Two dashed routes lead to A and B. A question mark signals that the destination is ambiguous.

**Narration:** This delivery robot can reach either station. But can the person tell which one it intends to visit? A clear route is not enough. Good human robot interaction makes the goal understandable, and makes intervention usable.

## 00:16–00:37 · Show the goal first

**Visible mechanism:** The destination label names B, and a cyan route with a direction arrow appears before the cart follows it. The lower card changes from known goal / expected path to seen motion / infer goal.

**Narration:** Show station B and the intended route before moving. Predictability asks whether motion matches an expected path to a known goal. Legibility asks whether an observer can infer the goal from motion. Neither proves safety. Here, a destination label and route preview reduce the need to guess.

## 00:37–01:00 · More than a red light

**Visible mechanism:** A color-only red circle is replaced by a labeled Cancel task button with a square symbol. A dashed focus outline, Enter key and traveling token illustrate keyboard activation. A separate text card says cancellation requested.

**Narration:** A red light alone excludes people who cannot distinguish its color. A spoken warning alone can be missed. Pair clear words and symbols, provide keyboard controls, and expose changing status to assistive technology. Test with the people who will use the system. Extra channels are not a substitute for testing.

## 01:00–01:24 · Requested is not stopped

**Visible mechanism:** A request token moves from person through supervisor to server. New dispatch becomes blocked immediately. A return token acknowledges acceptance; CANCELING appears. The cart continues along its track, then decelerates while a separate motion bar shrinks.

**Narration:** Watch the handoff. The person requests cancellation. Our supervisor blocks new task commands immediately, but the current motion still needs to stop. The action server accepts the request and enters canceling. The robot decelerates. A cancel response is not a canceled result, and neither message alone measures physical motion.

## 01:24–01:46 · Check action AND motion

**Visible mechanism:** The action card reports CANCELED for goal 1. An old zero-speed observation leaves the supervisor at MOTION UNKNOWN. A fresh zero-speed observation changes this to HELD, but dispatch remains blocked. A persistent warning distinguishes software cancel from an E-stop.

**Narration:** In our simulation, held requires a canceled result for this goal and fresh zero speed evidence. If feedback goes stale, display motion unknown and keep dispatch blocked. Do not paint a reassuring stopped label. This software cancel control is not an emergency stop or a certified protective system.

## 01:46–02:10 · Do not resume by accident

**Visible mechanism:** The held cart stays still while scene version 2 and plan version 1 mismatch. Recheck aligns both at version 2. Confirmation then changes to YES; the cart moves under new goal 2. An old goal-1 token is labeled IGNORE.

**Narration:** After intervention, this lab requires a fresh state check and deliberate confirmation of the current goal. Releasing a control does not restart the robot. A changed scene invalidates the old plan. Recheck, confirm, then send a new goal. Do not reuse a late result from the previous goal as evidence for the new one.

## 02:10–02:32 · May it start again?

**Visible mechanism:** Three cards report CANCELED, feedback age 2.0 seconds versus a 0.5-second limit, and confirmation NO. Name both missing conditions. Five quiet seconds are provided in the video; take as long as needed in this transcript.

**Narration:** Your turn. Cancellation finished. The last zero speed report is two seconds old. This lab allows at most half a second. The person has not confirmed a restart. May the robot move? Name both missing conditions before the answer.

## 02:32–02:52 · Keep dispatch blocked

**Visible mechanism:** The answer identifies fresh motion evidence and deliberate confirmation. Dispatch remains blocked until those checks are treated as satisfied in the illustrated continuation. NEW GOAL appears and the cart moves again. No real hardware is controlled.

**Narration:** No. Get fresh motion evidence and explicit confirmation after the state checks pass. Make intent visible, intervention accessible, and restart deliberate. Try the event trace in the AI Robotics lesson, and follow Humanoid Hub for more mechanisms explained.

**Untimed retrieval:** A two-second-old report exceeds the half-second limit, and restart is not confirmed. Either failure blocks dispatch. Obtain fresh evidence, recheck the current scene, and deliberately confirm the current goal; do not substitute a protocol response for measured physical state.