Frequently asked questions#
General#
Does Quantum Mind need a quantum computer? No. Everything runs on an ordinary computer: the quantum-like models and COBRA are classical code, and circuits run on a built-in NumPy simulator. A quantum computer is only needed if you choose to send circuits to IBM Quantum or Amazon Braket.
Does “quantum-like” mean the brain is quantum? No. Quantum-like models borrow the mathematics of quantum probability because it describes certain patterns in human answers (order effects, context effects). They make no claim about physics in the brain. See concepts.
Is there a quantum advantage? None is claimed. Every quantum-like model, including COBRA, is compared with a classical baseline, and the documentation reports when the classical method wins, which is often.
Is any of this validated with real robots and real people? Not yet. The human data are published survey aggregates; all robotics results use simulated people. See validation status for what is and is not validated, and how to run a pilot.
Installation#
Which Python versions are supported? Python 3.10 to 3.12, tested in CI.
Which extras should I install?
Command |
Adds |
|---|---|
|
the core library (NumPy, SciPy) |
|
Matplotlib and Plotly viewers, widgets |
|
Qiskit, Aer and IBM Runtime for circuits |
|
IBM Runtime and Amazon Braket |
|
Gymnasium environments |
|
Gymnasium and py_trees |
|
everything above |
I get ImportError: install quantum-mind[viz].
The plotting back ends are optional. Install the extra named in the message.
Qiskit fails to install on my machine.
Qiskit publishes wheels for common platforms; on older systems upgrade pip first
(python -m pip install --upgrade pip). The library works without Qiskit; only circuit export and
hardware runs need it.
Windows? The pure-Python parts work on Windows; the ROS 2 node follows ROS 2’s own platform support.
Using it#
How do I run on IBM Quantum hardware?
Save your IBM Quantum API token with qiskit_ibm_runtime.QiskitRuntimeService.save_account(...), then
call quantum_mind.circuits.run(circuit, 'ibm:<backend name>', shots). See
hardware. Hardware results are not included in the package.
How large can the simulated circuits be? At most 22 qubits; up to about 10 qubits and a few thousand samples train in seconds to minutes.
How do I fit my own data?
Put your counts in a dictionary keyed by condition and call quantum_mind.compare([...models...], data).
Tutorial 1 shows the format; each family’s page gives its design.
How do I use it on a robot?
The decision functions are plain Python and middleware-free. HumanModelService wraps them with
JSON-friendly messages, and integrations/ros2 provides a ROS 2 node. See tutorials 12 and 13.
How do I cite it?
Use the citation file (CITATION.cff) or the citing page.
Documentation#
Are the numbers in the tutorials real? Yes. The tutorials are executed when the documentation is built; every printed number and figure is the output of that run. The text states results in words, and an assertion in the tutorial checks each stated claim, so the build fails if a claim stops being true.
Why are some results worse for the quantum method? Because that is what the experiments show. Reporting it is deliberate.