Inside Glacier's Mission to Modernize Recycling

Engineering & simulationGlacier's proprietary AI modelmassively increased the rate of successful picking

The AI model analyzes visual data to detect item topography and identify flat or smooth surfaces for suction cups, allowing the robot to successfully pick open clamshell containers that previously failed because the suction point was on the hinge.

“our robot now is actually capable of not only detecting that an item is a correct target item, but it can actually kind of see the topography of that item and look for flatter or likely smoother surfaces just to constantly pick off of. So that's massively increased the rate of successful picking we were able to do.”

Engineering & simulationGlacier's AI vision system

The AI system uses height detection data to identify hazardous objects colliding with the robot's internal parts, triggering a software-controlled 'hiding' maneuver where the robot arm moves aside to avoid damage.

“what we did to counteract, that was also a kind of clever software solution, which is to say, now that we have height detection, We can also tell when something might collide with our internal parts. If that's the case, we can actually teach our robot to go and do what we call hiding, which is to move off to the side in a safe place for a 2nd or 2, let that object pass and then resume its activities.”

Engineering & simulationGlacier's AI model

The AI model is being trained on data from diverse facility inputs (clutter, throughput, contamination levels) to create a generalized system that can identify and sort recyclables across different Material Recovery Facilities without extensive reconfiguration.

“we're making huge strides towards a fully generalized AI model such that you can just kind of turn it on and it works out of the box, so to speak, but I think given the huge amount of data that's needed to get there, We're really fortunate to have each new deployment to learn from.”