Applications#
quantum_mind.applications.robotics: robots that ask people questions#
Name |
Purpose |
|---|---|
|
Three question pairs (object clarification, trust and hand-over, preference elicitation) with quantum-like, anchoring and Bayesian populations (illustrative parameters) |
|
Questioning designs: fixed order, reversed-order probe with a model, split order (model-free) |
|
Six hand-overs, with the trust question at the end only or also midway |
|
Competing human models weighted by BIC; predicted answer distribution plus uncertainty (entropy and model disagreement, in bits) for use as a confidence signal |
|
Decision rule: ask when the human models disagree or when the expected cost of a wrong action exceeds the cost of asking |
|
Intent from ambiguous commands and context cues; cues as Kraus operators rotated by an incompatibility angle, so cue order matters; every angle 0 gives exactly Bayes’ rule |
|
Answers in, prediction + uncertainty + ask/act decision out (JSON-friendly dicts); wrapped by the ROS 2 node in |
Findings from the companion simulation study (“Order-Aware Human Models for Robot Questioning and Trust”): splitting the question order beat model-based correction for recovering unprimed answers; the quantum-like model was the best predictor only when people followed it and were alike; asking about trust midway changed final trust under open-system dynamics (−0.076) and needed about 1,600 people per condition to detect. Use the ensemble rather than a single model.