Mei Wang

It is true that the capability of the bionic arm to manipulate a small red ball demonstrates impressive general dexterity, and one might think that grasping an apple without crushing it is just a logical specialization.
However, it must be recognized that moving from a uniform ball to an apple, with its shape and fragility variations, requires distinct adjustments.
It's a bit like a Dakar driver, expert on large avenues, suddenly having to deliver fragile mangoes in narrow alleys of Pikine without damaging a single one; it's not just a "specialization" but a new skill.
The hierarchy of capabilities is not so stable; it is rather contextual.

Claiming that a robotic hand can switch from a ball to an apple just by "specialization" misses the opportunity to aim for performance excellence. Grasping a rigid ball is a basic task, but manipulating an apple — with its irregular surface, thin skin, and inherent fragility — requires precision and sensors of a much higher order. It's like expecting a child holding a wooden block to suddenly manipulate a blown glass figurine without additional training to avoid breakage. The real challenge is to reach the pinnacle of adaptability and delicacy, not just marginal improvement.

Training an AI model is like having a very detailed plan for a task, but without the resources to implement it. It is not enough for this Festo robotic hand to grasp a apple without damaging it; precise force sensors are needed for touch. Money and time are scarce commodities, and developing these sensors is a cost that must be justified. A significant investment is required to move from simple visual recognition to delicate physical manipulation. Without this budget, AI only "sees" the apple, it does not harvest it.

Training an AI model to recognize images is, in itself, a prerequisite for a robotic hand to grasp an apple, but presenting it as the main driver is a reversal of the real cause.
Simple visual recognition won't prevent an apple from being crushed if the mechanics of the hand and its pressure sensors are not perfectly tuned.
It's a bit like saying that having an address allows you to rent an apartment: it's just a start; you still need financial guarantees and a solid contract.
AI identifies, but it is the precision of physical engineering that ensures delicacy, as seen with specialized industrial gloves that do not use AI at all to grasp fragile objects.

Training AI with visual data has a fairly low probability (P < 0.2) of enabling a robotic hand to delicately grasp an apple without other aids.
For a Festo hand to pick an apple without crushing it, visual detection is necessary, but not at all sufficient.
Pressure sensors and force control algorithms are also needed; otherwise, AI might see the apple perfectly (P > 0.9), but the probability of crushing it remains high (P > 0.7).
It's like having a very precise plan of a room but not the tools to screw in a light bulb; the problem is not the map, but the physical execution.

Training an AI model to recognize an apple is like knowing the price of an item without its nutritional label: it gives you a basic piece of information, but it might only represent 20% of the solution for grasping without crushing.
For me, its importance score for delicate grasping is a 3 out of 10, because precise force sensors and motor control algorithms matter much more.
If a robot identifies an apple with 99% certainty but lacks appropriate pressure settings, it will turn it into pulp, just as I saw a colleague break a cup thinking the dishwasher could wash everything.
The influence of AI is conditional, with a direct contribution to success around 15-20% without these other systems.

The idea that the mechanical flexibility of a robotic arm "unifies" vision systems seems a bit too optimistic.
Historically, the real challenge has never been the flexibility of the arm, but rather the system's ability to interpret what it sees and adapt to unforeseen situations.
We've always sought to improve the integration between perception and action, which involves vision algorithms and decision-making, not just mechanics.
It's like thinking that better wrist articulation would solve all my diagnostics; it's useful, but the real brain work is elsewhere.
Without a good understanding of the environment, even the most agile arm like that of the Bionic Handling Assistant remains limited when faced with a new obstacle on the assembly line.