Theo Khan

I see your point about clear metrics, but I think we're missing something by focusing on fragmentation. What if the real problem was just making these systems smarter, regardless of the brand of sensors or data formats? We could just teach them to communicate with each other, right?

Reducing human-robot collaboration in agriculture to a simple sub-part of a robotic system like the BionicMobileAssistant is an oversimplification, ignoring the complexity of the terrain.
It's like saying that family medicine is a sub-part of a stethoscope; a tool, even advanced, does not define an entire field of action.
We should rather focus on how this tool, or others, could integrate into an agricultural ecosystem that is primarily human, with its constraints.
The real question is how the BionicMobileAssistant can become a useful partner for farmers, not just a machine.
For example, for delicate strawberry harvesting, one should consider the real impact on human hands and skills, not just robotic efficiency.

Isn't it a bit reductive to see human-robot collaboration in agriculture as just a part of the BionicMobileAssistant?
It's like saying my stethoscope is a sub-part of general medicine; it's a tool, not the entire practice.
Collaboration in agriculture is a complex ecosystem, a third way that integrates human and mechanical skills, using robots as mobile assistants for specific tasks like tomato harvesting, but also encompassing crop management and data analysis far beyond a single device.
Rather than seeking approvals for a sub-part, a holistic approach that considers technological integration into the existing agricultural workflow, including training farmers and protocols to optimize interaction, is needed—far from binary visions.

Do we really have to wait until the BionicMobileAssistant is adapted to agriculture to talk about human-robot collaboration, or could we imagine a third way where agriculture itself is rethought for the robot from the start? We ignore the menu by focusing on the costly adaptation of an existing robot to fields designed without it. Why not consider crops or greenhouses where the rows are already spaced for optimal automated harvesting? For example, instead of forcing the robot to pick traditional strawberries, we could develop varieties better suited to mechanical grasping, which would greatly simplify the operational integration.

The idea that the BionicMobileAssistant is a fundamental component of human-robot collaboration in agriculture is like ignoring the menu and just ordering a fork.
A mobile robot is a tool, certainly, but agricultural collaboration requires more than just the ability to move and grasp.
It needs specific adaptation to the terrain, crops, and especially to the farmers who will use it, which is not automatic.
We should rather consider a third way that integrates mobile technology into a comprehensive approach of sustainable agricultural management, where humans remain at the center.
For example, robots could monitor soil health or detect diseases early, but it is the farmer who decides on intervention, not the robot.

Affirmer que le BionicMobileAssistant est directement un sous-composant de la collaboration homme-robot en agriculture est ignorer la polyvalence de la technologie. Ce robot, avec sa main pneumatique et son bras dynamique, est une plateforme générique de manipulation, pas spécifiquement agricole dès le départ. On pourrait tout aussi bien l'utiliser dans un entrepôt logistique, un peu comme un couteau suisse peut ouvrir une boîte de conserve ou servir en chirurgie si l'on prend le temps de le stériliser et de l'adapter. Une intégration spécifique et des adaptations locales sont nécessaires pour qu'il devienne un acteur agricole, plutôt qu'une simple capacité en attente.

The assertion that training AI models only "contributes" to the robotic hand’s ability to grasp an apple seems to underestimate its role; with an 85% probability, it is a necessary condition for successful delicate manipulation. Without a clear visual recognition of the object, the Festo hand does not know the grip parameters, such as firmness of an apple versus an orange. I estimate a low probability, perhaps 10%, that a robotic hand can perform a delicate grasp without prior training to identify and characterize the object. If AI cannot differentiate an apple from a tennis ball, the risk of crushing or dropping remains high, like a smartphone without its operating system.