Fusion-weight allocation must be predictable rather than opaque
Aliases: authority allocation · assistance gain · adaptive shared control
What it is
Predictable control authority allocation makes the relative influence of human and machine follow understandable conditions. Weight may adapt to risk, confidence, or phase, but a person needs to know when correction strengthens, input passes through, and why identical commands differ.
Why it happens
Operators learn dynamics from input–output relationships. Opaque weight changes make that mapping non-stationary, prompting exaggerated command, probing, or disengagement. Changes coupled to visible risk, boundaries, or confidence can be attributed to context.
Studying it
Fixed, rule-based adaptive, and opaque adaptive weights can vary change rate and cueing while measuring output prediction, compensation, conflict, transfer, and agency. Actual weight and rationale need logging. Mean performance can hide rare unpredictable overrides.
Where it stops holding
Exact formulas need not be exposed; behavioural boundaries and cause may suffice. Emergency constraints can immediately dominate but need explanation. Predictable does not mean fully user-configurable when safety is at stake.
Applying it
- Bind weight changes to few visible conditions and show the direction of current human/machine influence.
- Preview major override through haptic, visual, or resistance cue and retain the cause afterward.
- Test output prediction across contexts, monitoring overcompensation, repeated command, and abandonment.
Related
- Same group: X4.04.1 In shared control, human and machine inputs are fused into one output · X4.04.3 Arbitration of human–machine intent conflict needs timely feedback · X4.04.4 Shared control depends more strongly on sustained mutual trust than either full manual or full autonomy
- Adjacent: X4.01 Levels of autonomy · X4.05 Takeover and handoff
- Search terms:
control authority allocation·adaptive shared control·assistance gain