Ren Cohen

C'est bien ça le problème, on a déjà investi tellement là-dedans qu'on ne voit plus les défauts évidents. Si on avait su dès le départ que le système n'avait pas de retour haptique, est-ce qu'on aurait continué à dépenser des millions sur la reconnaissance visuelle de l'IA pour qu'elle puisse repérer une pomme qu'elle allait écraser ? On a besoin de voir les capteurs de force sur la main FESTO, avec un vrai chiffre en Newton, avant de continuer. Sinon, on jette l'argent par les fenêtres, comme d'habitude.

Yes, that's true, a Festo robotic hand that holds an apple without damaging it changes the game to see if AI training has really been useful. If we hadn't already invested all this money in AI, we would first check if the parts are up to par, like a plumbing quote before redoing the entire bathroom. Without that, we would wonder if we invested in image processing just to crush fruits.

There. This processing delay of AI is the problem, not the detection itself. If learning from images only adds a link to the chain, we have just one more reason to review the entire project before investing more funds. We shouldn't continue if automation of decision-making isn't instantaneous, regardless of how many photos we've processed through the system.

We could have wondered if we should really limit ourselves to factories and assembly, given all the money already invested in developing these hands. But if we incorporate the idea that it can also be used to manipulate hot objects or fine glassware, as you mentioned, then suddenly, we see that the scope broadens and justifies not throwing everything away. We would then need to review our safety protocols to include these new applications, not just think about microcircuits.

We should take a close look at this robotic hand, especially if it is capable of identifying different types of materials as you say. If we hadn't already invested money in it, would we think that the BionicSoftHand 2.0 is the only option to reduce errors, or would we look into what the competition is doing for better value?

Recognizing an apple with AI is one thing, but grasping it without crushing it is another, which depends on a precise calibration and mechanical engineering far beyond simple visual identification. If we haven't already invested so much in recognition, would we reallocate these resources if the robotic hand doesn't have sensors for delicate manipulation? It's like a car with a super GPS but no brakes: it knows where to go, but it ends badly. An AI model can "see" a red apple perfectly, but without well-tuned pressure sensors, this apple will turn into applesauce.

If we hadn't already invested so much in the idea that AI would solve everything, would we consider training a model as the essential condition for a robotic hand to hold an apple without crushing it?
The real challenge lies in sensor mechanics and actuator precision, far more than in visual recognition alone; if the hand doesn't "feel" the apple correctly, it will crush it.
Without proper engineering for graded force and tactile feedback, even the best AI cannot prevent a catastrophe.
Training AI is a facilitator, yes, but not a sufficient guarantee without these other pieces of the puzzle.
A robot can recognize an apple, but if its fingers are too rigid or its pressure sensors only have an 'on/off' mode, the apple will end up as compote.

The idea that AI training is just a "contributor" to the robotic hand’s ability to grasp an apple makes me pause; what's the point of an ultra-sophisticated hand if it doesn't know what to grasp or how to identify it? If we hadn't already massively invested in object recognition, would we launch a new delicate manipulation project without this foundation? It's like having a state-of-the-art vehicle but no road map: the mechanics are perfect, but the goal is lost. Without AI perception, the robotic hand couldn't even differentiate an apple from a tennis ball, let alone adjust its pressure. Think of AI as the brain guiding the muscles of the hand.

If we had to design this robotic hand to grasp an apple today, with no prior investment, we would really question whether AI training is the first thing to fund. It seems we're clinging to AI because we've already spent time and money on it, while the real problem remains the mechanics of the hand and its sensors. If the hand lacks the delicacy to feel pressure, even the best visual recognition would only identify the apple before it ends up mashed. Instead of seeing AI as the miracle solution, we should ask ourselves: Would it work without these physical sensors, regardless of AI power.

Claiming that the BionicSoftHand 2.0 is an example of human-robot collaboration in industry is a bit rushing.
If we started from zero, without the sunk cost of the general idea of collaborative robots, would we classify a simple robotic hand as a full collaboration?
In reality, a hand, even very sophisticated, remains a component; it only becomes "collaborative" if integrated into a larger system, as if saying a forklift is a complete warehouse.

What is at stake is the very ability of the hand to grasp without destroying. The breaking point would be if the robotic hand didn't have truly effective pressure sensors at the fingertips, even if AI perfectly identifies the apple. Without this physical feedback, AI couldn't prevent the apple from turning into mush, regardless of its "training".