A 7-DOF manipulator with sensorized fingertips that picks heterogeneous SKUs out of mixed totes at human-equivalent throughput — and improves over time.
A regional 3PL handling 40,000 SKUs needed a bin-picking solution. Existing systems on the market struggled with deformable items, glossy packaging, and sub-second cycle requirements. They also couldn't be retrained without vendor visits.
CrateHand v3 combines a custom 7-DOF arm, optical-tactile fingertip sensors, and a vision-language-action policy that grasps from natural-language item descriptions. It self-improves: every successful pick augments the on-device dataset, every failure is reviewed overnight and pushed as a model update.
Three months post-deployment, throughput is 1,140 picks/hour with a 99.4% success rate across the full SKU range. The system has retrained itself for 3 new product categories without engineering involvement. ROI broke even in month 5.
Most of our deployments start as a conversation. Tell us about the environment, and we'll tell you whether this is the right platform — or what to build instead.