A warehouse selling spare parts may add a new item every day. Teaching a robot a motion for each box can become a job in its own right. Sereact builds Cortex around a different premise. The robot sees what arrived, relates it to the task and selects a way to handle it. The important buying question is how far that generalisation extends across the assortment that actually reaches a particular station.
- 01The model Sereact describes a vision-language-action system complemented by a model that predicts the consequences of possible movements.
- 02The equipment A single-arm cell picks and checks items. A dual-arm workstation can open packaging and handle returned goods.
- 03The evidence The company names installed systems and publishes model experiments. The two sources answer different questions about a proposed warehouse deployment.
01 / ProductWhat Cortex contributes to a robot cell
In Sereact's picking system a tote arrives at a conveyor or storage port. The cell identifies the item, plans a grasp and adjusts its motion as the pick proceeds. A quality check called Lens can flag damage or a wrong item before final placement. The destination could be a carton, a tote or an outbound conveyor. A facility therefore buys a workflow with inputs and outputs, even when the underlying software model is what first caught its attention. Single-arm picking sequence
The Cortex 2.0 technical description explains one way Sereact evaluates possible actions. A world model predicts possible future scenes. The system scores those alternatives for progress, stability and efficiency, then uses the result to guide execution. Predicting what an unstable grasp might look like gives the policy information before an object actually slips. This account describes that documented generation of the model. The company's landing page also carries a later Cortex version label. Technical explanation Company overview
Returns make the distinction between movement and judgement especially clear. Opening a shoebox, lifting out its contents and reading a label are physical operations. Deciding whether that pair can go back on a shelf draws on the retailer's own rules. The returns page describes opening, scanning, inspecting and sorting with a dual-arm station. A retailer evaluating the station still needs to define which conditions qualify for resale, refurbishment or a person's inspection. Returns-handling system
02 / AudienceWho has an assortment worth testing
One promising audience is an operator that already delivers source totes reliably but still struggles to staff the picking face through busy periods. Its records can identify the most frequent lines, the awkward packages and the arrivals with inconsistent labels. Those are valuable ingredients for an evaluation. If the station frequently waits for stock, look at the upstream supply of totes as well. Faster grasping alone cannot fill a carton from an empty source.
Another audience is a retailer whose returns mix changes with every promotion. Inbound orders arrive with inventory records and expected quantities. A return might arrive in the wrong box or with its packaging open. A demonstration with clean, unopened retail packs tests relatively little about that situation. Include confusing but harmless cases such as a different colour in a correctly labelled box or a damaged carton containing an intact product. Specify which of those differences has a financial consequence.
03 / WorkflowA proposed evaluation from first view to confirmed placement
Start with an ordinary output from the warehouse management system, such as the list of picks on a busy afternoon. Group the work by physical properties as well as by product identity. A plastic bag containing a rigid part can behave differently from a loose bag of soft goods, even if both carry the same shipping weight. Keep a portion of the assortment unseen until the evaluation. That gives the claim about unfamiliar items a concrete meaning the operator can check.
For each attempted pick, record the intended identity, the item actually placed and any exception. Count a successful retry separately from a first-attempt success. A reset carried out promptly by an operator can be an acceptable business result, but its staffing cost disappears if the test logs only the final successful placement. Track the time between a failure and the point at which the following order can move again.
Run a second, smaller test on returns. Give the team paired examples with the same label and different physical conditions. Compare the robot's proposed destination with the disposition that a trained employee would choose. Where employees themselves disagree, write a clearer classification rule before attributing the disagreement to the model. A minor mark on a shipping carton and a defect on the item may require different action even though both appear as damage in an image.
04 / PricingHow to scope a commercial proposal
Sereact invites prospective customers to book a demonstration and discuss a pilot with its technical team. The public contact route does not state a universal price per robot, per pick or per software licence. Sereact lists robotics-as-a-service as a feature of its returns offering. CEVA Logistics also describes delivering returns handling through that model. A reader should obtain the terms for their own region, cell, interface and operating hours. Pilot enquiries Partner returns model
| Work being scoped | Public evidence | Item to obtain in writing |
|---|---|---|
| Pick cell | Single-arm station with Lens inspection | Eligible item mix, integration and acceptance test |
| Returns service | Sereact lists robotics-as-a-service for this offering | Contracting party, unit of service and exceptions |
| Pilot | Technical team discusses trials and demonstrations | Equipment, site preparation, duration and support |
Sereact's pilot and commercial contact and returns offering, consulted 2026-11-21.
Use the number of accepted order lines as one denominator. Dividing a quote by every arm movement rewards a system for retrying. Dividing by all outbound boxes can hide manual lines in orders that also contain automated picks. In a returns project, use completed, correctly classified items and record unresolved dispositions. Ask the supplier to describe how a planned software update or an unavailable upstream conveyor would affect any performance commitment.
05 / DistinctionsWhat the model evidence and customer examples each show
Sereact's technical article includes tests on a dual-arm setup and reports that considering more possible futures can improve decisions at the expense of time per step. A shoebox task is one of its examples. It helps explain why extra planning can pay off when one poor move disrupts the rest of a sequence. The experiment belongs to that model and apparatus. A customer should collect separate evidence on its own incoming containers, staff handoffs and output standards. Cortex experiments
Its case library shows picking at Monta and Active Ants, including work with AutoStore, alongside grocery and retail examples. These are useful starting points for questions about presenting totes to a cell and coordinating with existing equipment. For comparison, Mujin covers industrial robot control and the planning of material flows. Ambi Robotics offers another perspective when sorting packages or stacking them downstream is the central job. Sereact deployment library
06 / QuestionsThe failures worth discussing before a rollout
A visibly similar item is one of the hardest kinds of error to notice in an upbeat demonstration. A robot may grasp it cleanly and place it in exactly the right carton. Ask what happens when the scanned identity disagrees with the expected one, how the operator sees the discrepancy and which system has authority to release that carton. A pause with an interpretable reason can be preferable to a smooth movement that sends the wrong part to a customer.
Updates raise a second operational question. Sereact says it validates updates before releasing them. An operator can turn that into an acceptance conversation by keeping a set of difficult packages from the original trial, including examples of costly misclassification. Confirm how those are checked when packaging changes or software is updated. The exception desk should know how to identify a version-specific problem as well as how to clear the immediate blocked order. Returns update process
07 / DecisionChoose a station and a measurable outcome
The most informative first project is a bounded operation with a clear owner at each end. Use an outbound pick when the upstream tote and expected item are known. Use a return when the operator can supply disposition rules and independently inspect the result. The scope then gives both warehouse staff and the vendor a fair way to interpret what the equipment achieved.
Order fulfilment
Measure correctly completed lines and the time lost to retries on a held-back assortment.
Retail returns
Write the resale and escalation rules with the people who approve the stock adjustment.
Automation integration
Map the upstream tote arrival, the Lens exception and the downstream confirmation to named systems.
A business worth understanding.
Suggest your business or one you find interesting. Tell us what you want to understand about its product, positioning, design or workflows.
Suggestions are free. Selection and publication stay with the desk.
- Cortex 2.0 technical overviewConsulted
- Single-arm pick and placeConsulted
- Dual-arm returns handlingConsulted
- Warehouse case studiesConsulted
- Contact and pilot enquiriesConsulted
- Sereact company and deployment overviewConsulted

