Answer from AI Robotics Australasia. .
How a bin-picking cell works
A 3D sensor scans the bin and produces a surface the software can reason about. A grasp planner chooses a part and a grip that the gripper can reach without hitting the bin wall or another part. The robot executes that path, then the cell re-scans or updates the cloud and tries again.
Around that loop sit the unglamorous pieces: a gripper that matches the part, a way to singulate or re-grip if the first hold is wrong, and a rule for parts the system should skip. Cycle time includes the failures, not just the clean picks in a demo.
When it is the right problem to automate
Bin picking is justified when parts arrive jumbled and fixturing them by hand would cost as much as the operation you hoped to automate. It is common upstream of machine tending, kitting, and some depalletising tasks where the stack is not a neat pattern.
It is the wrong spend when a tray, a blister, or a simple fixture would present the part for a fraction of the cost. Vision does not get points for solving a presentation problem the packaging could have avoided.
What makes it fail
Shiny metal, black rubber, clear film, thin sheets, and parts that tangle are the usual hard cases. So is a bin that was packed with no thought for a robot's approach angle. A trial on your parts, in your bin, is the only evidence that matters.
We trial the real part before a cell is specified. If the cloud or the grasp is not reliable, the answer may be a presentation change rather than a more expensive sensor.
Related questions
What is random bin picking?
Random bin picking means the parts are jumbled, so each grasp starts from a new pose. The cell needs 3D vision and a planner. It is harder than picking from a tray where every part sits the same way.
What is the difference between bin picking and depalletising?
Depalletising removes cases or parts from a pallet, often in a known pattern. Bin picking removes individual parts from a container where the pose is not guaranteed. A slumped or mixed pallet starts to look like bin picking.
Does bin picking need 3D vision?
Random bins do. A structured tray can often be picked with 2D vision because depth is already controlled by the tray.
How fast is a bin-picking robot?
Fast enough only after you count failed grasps, re-scans, and the parts it skips. A quoted picks-per-minute figure from an easy sample is not your rate. The trial on your part sets the number.