You say that an autonomous mobile robot like the BionicMobileAssistant does not guarantee human-robot collaboration if the human cannot intervene, and that a safety threshold is needed; but I wonder if simple manual intervention is enough to speak of collaboration.
The ease of understanding and modifying the behavior of an autonomous system, not just stopping it, is what really matters.
In Senegal, we see this clearly with our mobile services; even the best app, if not intuitive and easily reconfigurable for our changing realities, ends up frustrating users instead of helping them.
AI can enable smooth interaction, but true collaboration depends on the flexibility of the robot to adapt to humans, not the other way around.
For example, a farmer should be able to reprogram his spraying drone for a new plot without needing an IT degree, or it's not collaboration, it's rigid assistance.