Rehabilitation Robotics and Assistive Shared Control
Status: CLOSEST EXISTING ENGINEERING for the "stabilize, don't replace" principle. Foundation with critical scope limitation.
What they did
The assistive shared-control literature -- particularly smart wheelchair navigation and rehabilitation training -- independently articulated the core insight that over-assistance harms the human.
- Soh & Demiris (2015): in training/rehabilitation contexts, "the primary objective is long-term development rather than short-term task completion."
- Urdiales et al. (2008): design for the person to be "always in charge of his/her own motion"; explicitly warn that "excessive assistance may lead to loss of residual capabilities."
- Both cite learned helplessness (Seligman) as a design failure mode to avoid.
This literature formalized replace-vs-stabilize as a design philosophy: under-help produces frustration; over-help produces learned helplessness and skill atrophy; full takeover removes the sense of being in charge. The design response is "assist-as-needed" -- decomposed into "when-to-help" and "how-to-help" models.
Critical differences
- Scope: Their "capacity preservation" targets motor skill, navigation competence, training development. The leap to generalized self-directed action capacity is not made.
- State estimation: Based on user performance and skill, not neurobiological state or physiological monitoring.
- Understanding layer: None.
- Normative governance: No Mayeroff-structured caring layer. No formal objective function over human capacity.
- Generalization: Domain-specific (motor rehab, wheelchair navigation), not general-purpose.
Positioning statement
Assistive shared-control robotics independently articulated the "stabilize, don't replace" principle and formalized assist-as-needed paradigms in specific rehabilitation domains. This work generalizes the same principle from motor skill preservation to general capacity for self-directed action, adds physiological state estimation and AI understanding, and introduces caring governance. The rehabilitation robotics insight -- that over-assistance creates the very helplessness the system is meant to address -- is the engineering expression of the backfire prediction (P1).
Sources
- Soh & Demiris 2015 (J. Human-Robot Interaction 4(3):76-100)
- Urdiales et al. 2008 (AAAI, "An Adaptive Scheme for Wheelchair Navigation Collaborative Control")