Industrial robotics is a solved problem inside a controlled cell. Outside of it, everything is unknown. Lighting changes, floor markings are worn away, a pallet arrives rotated, a pedestrian walks through the aisle.
Autonomy is a stack, not a switch
A mobile robot that navigates reliably is a stack: localisation from lidar and cameras, a map that stays accurate as the environment shifts, a planner that respects the rules of the space, and a safety layer that can override all of it. Failures at any layer show up as the robot stopping, which is the correct behaviour.
Picking is where autonomy gets expensive
Structured pallets are straightforward. Mixed, damaged or partly transparent items are not, and handling them well means combining 3D perception with force sensing. The honest state of the art is good performance on the easy eighty per cent, and expensive effort on the rest.
Maintenance is the product
Deployments succeed when the team treats uptime as the deliverable: spare parts, a fleet dashboard, a recovery procedure when a robot is genuinely stuck, and a technician who can read its logs. The robot is the visible half of the product; the operations around it are the other half.
La robotique industrielle est un problème résolu dans une cellule maîtrisée. Au dehors, tout est incertain. La lumière change, les marquages au sol s’effacent, une palette arrive pivotée, un piéton traverse l’allée.
L’autonomie est une pile, pas un interrupteur
Un robot mobile qui se déplace de façon fiable est une pile : localisation par lidar et caméras, une carte qui reste juste quand l’environnement bouge, un planificateur qui respecte les règles de l’espace, et une couche de sécurité capable de passer outre tout le reste. Une défaillance à n’importe quel niveau se manifeste par un arrêt — ce qui est le bon comportement.
Le picking, là où l’autonomie devient coûteuse
Les palettes homogènes ne posent pas de difficulté. Les objets mélangés, endommagés ou partiellement transparents, si. Les traiter correctement combine perception 3D et détection de l’effort. L’état réel de l’art est bon sur les quatre-vingts pour cent faciles, et cher sur le reste.
La maintenance est le produit
Les déploiements réussissent quand l’équipe prend le taux de disponibilité comme livrable : pièces détachées, tableau de bord de flotte, procédure de reprise quand un robot est réellement bloqué, et un technicien capable de lire ses journaux. Le robot est la moitié visible du produit ; l’exploitation autour, l’autre moitié.
Industrial robotics is a solved problem inside a controlled cell. Outside of it, everything is unknown. Lighting changes, floor markings are worn away, a pallet arrives rotated, a pedestrian walks through the aisle.
Autonomy is a stack, not a switch
A mobile robot that navigates reliably is a stack: localisation from lidar and cameras, a map that stays accurate as the environment shifts, a planner that respects the rules of the space, and a safety layer that can override all of it. Failures at any layer show up as the robot stopping, which is the correct behaviour.
Picking is where autonomy gets expensive
Structured pallets are straightforward. Mixed, damaged or partly transparent items are not, and handling them well means combining 3D perception with force sensing. The honest state of the art is good performance on the easy eighty per cent, and expensive effort on the rest.
Maintenance is the product
Deployments succeed when the team treats uptime as the deliverable: spare parts, a fleet dashboard, a recovery procedure when a robot is genuinely stuck, and a technician who can read its logs. The robot is the visible half of the product; the operations around it are the other half.
Conversation (0)
Sign in to comment or reply.