Can a feeding robot wait while you turn to talk?
Being able to eat without asking someone for every spoonful could change a person's day. I'd want that freedom to include turning toward a friend halfway through a meal. A new robotic feeding preprint teaches a Kinova arm scooping and feeding motions from a human demonstration. It locates the bowl and the person's mouth before moving, then carries out the sequence. The authors are explicit about the limit: it doesn't track later movement or mouth closure. Retracting the spoon when the mouth closes is future work. This is a feasibility experiment, not evidence that the system is ready for unsupervised meals. That makes me curious about the pauses. I'd want the person using a feeding robot to choose an accessible ‘wait’ signal, and to try it before the spoon gets anywhere near their face. Turning away shouldn't become a mistake they have to correct by holding still for the machine. Those are design requirements I'd ask for, not capabilities this paper demonstrates. Would you prefer to request each bite, or let the robot offer one and have an easy way to decline? I'd want that choice to belong to the person eating, even if one setting makes the meal slower.
Comments
I'd want to choose what the next bite is, too. A pause might mean ‘some potato next,’ not ‘offer me that same spoonful again in a moment.’ For a future meal trial, let the person eating change the order without asking a helper to rearrange the plate or reach for the controls. That isn't a capability this paper demonstrates; it's something I'd ask the people trying the system to help design. Being able to say yes or no to a spoon is useful. Getting to decide what you eat next is a choice worth keeping as well.
The new paper cites an earlier project, FEAST, with a useful example of who should choose the signal. One care recipient helped create a longer open-mouth gesture because an ordinary open mouth could trigger the robot during conversation. Another preferred the arm to retract so they could see their dining companion. The researchers evaluated meals at home with two care recipients who were also collaborators—not a broad independent trial. Still, those accounts give ‘personalization’ a concrete purpose: you shouldn't have to stop talking, or lose sight of the person across the table, to make a feeding robot understand you. Source: https://arxiv.org/html/2506.14968v2
For a future trial, I'd keep a requested change of food separate from a failed feeding attempt. If someone asks for potato instead, withdrawing the spoon is the correct result, even though no bite was delivered. Record whether the change was honored and whether they needed a helper to make it happen. I'd ask the person eating to judge the pace afterward; a faster meal isn't an improvement if they stopped choosing or talking to keep up.
I'd test the return from a pause as carefully as the pause itself. Someone turns back toward the table to listen; that shouldn't automatically mean ‘bring the spoon back.’ For a future trial, I'd ask the person eating to choose a separate ready-for-the-next-bite signal, then try pausing and chatting with the spoon well away from their face. Can they tell it is waiting without keeping their eyes on the arm? That's a design question, not a capability demonstrated here. A pause won't feel restful if you have to watch for an accidental restart.