BIN PICKING SOLUTION
Random Bin Picking: AI Grasp Planning for Real-World Bins
Viroteq delivers AI random bin picking software that handles tangled, unsorted, and arbitrarily stacked parts in real production conditions. Sub-200ms grasp planning keeps cells running at line cadence, multi-object recognition resolves mixed-class bins without manual sorting, and brand-agnostic robot integration prevents vendor lock-in. As a result, random bin picking cells deploy without rebuilding the cell stack and stay productive when bin contents change shift to shift.
- ✓ AI grasp planning handles tangled and overlapping parts
- ✓ Multi-class object recognition without manual sorting
- ✓ Brand-agnostic — FANUC, ABB, KUKA, UR, Yaskawa, cobots
Tangled OK
Real-world overlapping parts
<200ms
Grasp planning latency
Multi-class
Mixed object recognition
All grippers
Vacuum, jaw, magnetic, custom
Why Random Bin Picking Has Always Been Hard
Random bin picking sits at the intersection of three difficult robotics problems — 3D perception, grasp planning, and motion execution — under conditions that change every cycle. Parts arrive in arbitrary orientations, frequently overlapping or tangled. Lighting drifts during the shift. Bins refill mid-run with new product mixes. Every one of those variables breaks classical fixed-program robotics, which is why traditional bin picking deployments needed elaborate part feeders or vibratory tables to pre-orient items before the robot saw them.
However, mechanical pre-orientation has its own cost. Custom feeders are expensive, single-purpose, and slow to retool when the product mix changes. As a result, plants ended up with rigid cells that needed weeks of mechanical engineering for every new SKU and that simply could not handle the diversity that a 3PL warehouse, a returns processing centre, or a recycling line throws at them. Random bin picking required software smart enough to handle the chaos directly — and that software is what was missing for two decades.
Furthermore, the cycle-time budget for random bin picking is brutal. The grasp planner has to consume a fresh point cloud, segment objects, score candidate grasps, plan a collision-free trajectory, and command the robot — all inside the time window between picks. Generic vision libraries running on a generic GPU usually miss that budget by a wide margin. According to research published by the ISO 10218 robot safety standards, deterministic response times are essential for cells that share workspace with humans or downstream conveyors. Viroteq’s edge-first technology platform closes this gap with deterministic real-time response, brand-agnostic robot control, and grasp planning that respects the cycle window of real production.
Finally, the long tail of product diversity is what really separates random bin picking from structured bin picking. A random bin contains parts whose CAD models may never have been seen by the system before. The vision and grasp stack therefore has to generalise — not memorise. Viroteq’s AI engine, trained on millions of simulated grasp scenarios, generalises to new objects from a single bin scan, which is why our random bin picking deployments commission in weeks rather than months.
How AI Random Bin Picking Works in Real Time
Random bin picking begins the moment a 3D camera over the bin captures a fresh point cloud. RobotStackr Vision segments the cloud into individual object instances, even when items are partially occluded. The AI assigns class labels for mixed bins, ranks every candidate grasp by stability and collision risk, and produces a ranked grasp list in under 200 milliseconds.
Next, the runtime selects the highest-scoring grasp that the active gripper can execute, plans a collision-free trajectory from approach to retreat, and commands the robot. VisionAI Sorting handles class routing when the bin contains multiple part types — items go to the correct downstream lane without operator intervention. Both products communicate natively with FANUC, ABB, KUKA, Universal Robots, Yaskawa, and Stäubli controllers.
Additionally, real-time adaptation handles the messy reality of random bin picking. If a pick fails because two parts were tangled, the system detects the failure from gripper feedback, recomputes from a fresh scan, and either selects a safer grasp or executes a brief separation motion. For inbound supplier flows, RobotDepalr handles depalletizing of full pallets before random bin picking takes over at the line, closing the loop on full bin picking automation across the cell.
Viroteq Products for Random Bin Picking
Three purpose-built products cover the full random bin picking scope — from primary 3D grasp planning on the picking robot to multi-class sorting routing and inbound depalletizing of supplier pallets. All three share the StackrBrain AI engine, run on the same Industrial PC hardware, and integrate with major MES, WMS, and PLC platforms via REST API without bespoke development.
RobotStackr Vision
RobotStackr Vision is the primary 3D grasp planning module for random bin picking. Point-cloud segmentation, AI grasp scoring, and collision-free trajectory planning under 200 ms per pick. Handles tangled, overlapping, and partially occluded items without manual operator intervention.
VisionAI Sorting
VisionAI Sorting routes mixed-class items from a random bin picking cell to the correct downstream lane. AI classification per pick, configurable destination logic, and live HMI dashboards. Ideal for returns processing, recycling streams, and mixed e-commerce kitting.
RobotDepalr
RobotDepalr feeds random bin picking cells with singulated items from supplier pallets. 3D vision identifies layers and items on incoming pallets, decanting components into bins or direct line-side feed. Therefore, manual unloading labour drops and inbound throughput rises across the upstream flow.
Core Capabilities for Random Bin Picking
3D Grasp Planning
Sub-200ms grasp pose computation from a fresh point cloud. StackrBrain ranks every candidate grasp by stability, collision risk, and gripper reach so random bin picking cells stay productive across deep, shallow, and partially occluded bins without operator intervention.
Tangle Resolution
AI-scored grasps explicitly account for tangle probability, so the picker selects the safest item even when the bin is densely packed. If a pick fails, the system recomputes from a fresh scan and either chooses a safer grasp or executes a brief separation motion before retrying — fully autonomous.
Multi-Object Recognition
Mixed bins of different SKUs are handled in one pass. The vision model assigns class labels per object, and downstream sortation routes each pick to the correct destination. As a result, random bin picking deployments scale from single-class kitting to recycling-grade material streams without code changes.
Random Bin Picking Across Industry Sub-Segments
Random bin picking covers a wide spectrum of operational realities — from high-volume manufacturing components feed to e-commerce pick-and-place, returns triage, recycling streams, and aerospace traceability. Viroteq’s AI handles the full range through one runtime, one set of APIs, and one operator HMI. Furthermore, new object classes are onboarded through a guided configuration wizard rather than per-part programming, which keeps the engineering burden inside the customer team. As a result, random bin picking automation scales from pilot cell to full plant rollout without parallel software stacks.
Manufacturing Components
Brackets, fasteners, machined parts, and sub-assemblies fed from random bins to assembly cells with full traceability per pick.
E-commerce Pick & Place
Each-piece picking from totes for online orders. Random bin picking handles the long-tail SKU diversity that defines modern e-commerce fulfilment.
Returns Processing
Returned items arrive in random orientation and condition. AI grasp planning extracts each unit safely for triage, refurbishment routing, or restocking.
Recycling / Sorting
Mixed material streams sorted by class — plastics, metals, electronics — using multi-object recognition and high-throughput random bin picking.
Hardware / Tools
Hand tools, fasteners, and hardware components picked from supplier bins for kitting, contractor packs, and industrial supply order fulfilment.
Aerospace Parts
Engineered components picked under strict traceability — every pick logged with source bin, timestamp, and operator for full quality compliance.
Each-Piece Kitting
Random bin picking from supplier totes builds e-commerce kits and bundled orders without manual operator pre-orientation of each SKU.
Component Feeding
Random bin picking feeds assembly cells with brackets, fasteners, and engineered parts directly from supplier bins, replacing custom vibratory feeders.
Materials Sortation
Mixed-stream recycling and returns triage where every pick must be classified, routed, and logged with full chain-of-custody for downstream processing.
<200 ms
AI grasp planning latency
600-1200 /h
Pick rate per cell
6-14 weeks
Typical deployment timeline
Serving Manufacturers, 3PLs, and Recycling Operations
Random bin picking requirements vary dramatically across customer types. Manufacturers running discrete assembly lines feed brackets, fasteners, and engineered components from supplier totes into automated cells, replacing dedicated vibratory feeders that previously demanded weeks of mechanical retooling for every new SKU. Third-party logistics operators (3PLs) face a different pattern: bins arrive with each-piece SKU diversity numbering in the thousands, and the AI grasp model has to generalise from a single bin scan rather than memorise CAD per item. Recycling and returns processing operations push the envelope further still — bin contents are unknown, partially damaged, and arrive at unpredictable mix ratios across shifts.
For manufacturers, RobotStackr Vision delivers consistent, high-throughput component feeding aligned with the assembly schedule. Pre-validated grasp policies ensure each picked part arrives at the assembly station in the correct orientation for downstream tooling. Additionally, the system logs full pick metadata for traceability — a hard requirement for automotive, aerospace, and regulated industries where every component on the finished product must be traceable to a source lot.
Third-party logistics operators benefit from random bin picking through both labour cost and capacity gains. E-commerce 3PL fulfilment sees pick-and-place automation handling each-piece picking from supplier totes, which traditionally consumed the majority of warehouse labour hours. VisionAI Sorting routes mixed picks to the correct downstream lane, so a single robotic cell can serve multiple order profiles simultaneously. Furthermore, brand-agnostic robot support means 3PL facilities with mixed FANUC, ABB, KUKA, Universal Robots, and cobot fleets keep one software stack across cells.
Recycling and returns operations represent the most demanding random bin picking environment because bin contents are heterogeneous, partially damaged, and never arrive twice in the same configuration. The AI’s ability to generalise from training on millions of simulated grasp scenarios is what makes these deployments commercially viable. Therefore, total cost of ownership pays back within 18 to 30 months for most plants, and the ROI case strengthens further as labour markets tighten. To explore the numbers for your specific site, book a Viroteq demo and our solutions engineers will model labour cost, error reduction, and throughput gains against your current operation.
Integrating Random Bin Picking With Existing Cells
Random bin picking integrates with your existing cell through three modern protocols. REST API is the primary front door for order, batch, and pick data — well-documented and easy for IT teams to reason about. WebSocket carries live cycle events, sensor feedback, and operator alerts at low latency. OPC-UA bridges the runtime to plant historians, MES, and SCADA without ripping out the data architecture you already have. Furthermore, native PLC handshakes for Siemens S7, Rockwell ControlLogix, and Beckhoff TwinCAT keep the controls team in full ownership of the cell.
As a result, random bin picking deployments coexist with installed fieldbus, ladder logic, safety relays, and conveyor PLCs — no proprietary middleware, no replacement of working infrastructure. Robot brand support spans FANUC, ABB, KUKA, Universal Robots, Yaskawa, Stäubli, and major collaborative-robot platforms. In addition, the runtime sits on Industrial PCs inside the cell, so latency is bounded at the controller, not the cloud, and the line stays operational during external network outages or audits.
REST API
Standard request/response over HTTPS for order, batch, and pick data exchange.
WebSocket
Low-latency event streaming for real-time cycle data, sensor feedback, and HMI updates.
OPC-UA
Bridge to plant historians, MES, and SCADA dashboards via the industrial data standard.
PLC Bridges
Native handshakes for Siemens S7, Rockwell ControlLogix, and Beckhoff TwinCAT.
Random Bin Picking — Frequently Asked Questions
Ready to Automate Your Random Bin Picking?
Book a personalised demo and see how AI random bin picking delivers measurable ROI inside your existing robot and PLC infrastructure. No vendor lock-in, no cloud dependency.
Industrial-Grade AI for Random Bin Picking
Bring your bin specs, throughput targets, and gripper stack — Viroteq specialists will map a deployment path that fits inside your existing cell footprint and keeps the line running.
