All Foundries

Commodity Management & Trading

Transforming trading ecosystems through data-driven insights and secure, agile platforms that empower global commerce.

Customer & Marketing Strategy

Reimagining customer engagement with personalized, connected experiences that build trust and long-term value.

Digital Engineering

Turning ideas into scalable digital solutions—accelerating transformation through design, technology, and engineering excellence.

Station Studio

Redefining mobility hubs with digital innovation to create smarter, sustainable experiences for people and businesses on the move.

Supply Chain & Industry 5.0

Building resilient, intelligent supply networks that anticipate change and deliver efficiency with purpose.

Sustainability

Driving responsible innovation to help organizations achieve net-zero goals and create a positive impact for future generations.

Physical AI

Physical AI gives machines, robots and physical operations the ability to perceive their environment, decide what matters and act in real time. Factories, plants, stores and infrastructure become systems able to sense conditions, learn from them and improve on their own, closing the gap between the physical and the digital world.

Next-Gen Infrastructure

From networks as infrastructure to connected platforms that enable critical, autonomous and intelligent operations.

Retail

From satisfaction surveys to EBITDA: closing the loop on customer pain points

A department store retailer in Mexico turned a biannual NPS programme into a closed-loop customer listening mechanism, connecting feedback to pain-point initiatives and incremental EBITDA.

From satisfaction surveys to EBITDA: closing the loop on customer pain points

Retail

Unified returns management

A retailer processing more than two million returns a year moved from channel-specific workflows toward one digital returns system. The initiative connects request creation, status tracking and routing across sales and return channels.

Unified returns management

Contact

Vision-Based Robotic Picking

Small, irregular metal components are difficult to pick reliably at industrial scale. For a manufacturing client, NTT DATA designed an end-to-end, vision-based pick-and-place solution combining pose estimation, grasp strategy, cycle control, high-fidelity simulation and lab validation. The initiative integrated and tested the solution in a laboratory setting, with a single-robot architecture as the target industrial design.

A vision-guided industrial robot picks irregular metal components from a blue bin inside a precision manufacturing cell.

Single-robot design

Two handling steps in one cycle

One-shot adaptation

Faster path to new variants

Simulation-first

Virtual validation before integration

Reusable foundation

Twin and control assets for scale

A single industrial robot transfers metal components between input, processing and output stations in a compact pick-and-place cell.

Coordinated vision, grasping and motion for flexible robotic handling

Bringing vision, grasping and motion control into one coordinated cycle, so a robotic cell can handle greater part variability with a clearer path from lab validation to industrial deployment.

A two-robot process and highly variable metal parts made reliable picking, cell simplification and industrial rollout inseparable engineering challenges

Objectives

The initiative set out to industrialize a scalable pick-and-place process for small, highly variable and irregular metallic components. This required robust pose estimation, reliable grasping and coordinated task execution, while redesigning the cell around a single robot and validating its behavior virtually before shop-floor integration.

Opportunity

With the lab-validated simulation environment, control software and parameterized model-tuning approach in place, the client has a foundation for onboarding additional part references without re-engineering the complete process. The next opportunity is to extend the architecture across further product families, confirm performance under production conditions and use measured operating data to refine cycle time, stability, maintenance and quality outcomes.

Turning part variability into a repeatable robotic cycle

The challenge went beyond recognizing an object. The cell had to perceive, grasp and move irregular components consistently, while reducing architectural complexity and limiting deployment risk.

Key Challenges

Top-down view of varied machined metal parts in a blue bin, with a depth camera and robot wrist positioned for vision-based picking.

1: Perceiving irregular components
Small metallic parts arrived in different shapes, orientations and levels of occlusion. The initiative required robust pose estimation that could adapt to new variants without starting each model from scratch.

Close-up of a robotic gripper aligning with a complex metal component on a precision assembly fixture as a technician observes.

2: Engineering a dependable grasp
Pose alone was not enough. The gripper strategy had to accommodate complex geometries, including a slider body and puller, and support more stable handling across the cycle.

Two industrial robot arms share a tightly configured manufacturing cell with duplicated equipment, guarding and cabling.

3: Consolidating two robotic steps
The existing flow used two robotic arms for the full activity. The target architecture needed one robot to execute both steps, with coordinated task logic and cycle management.

Two robotics engineers validate a pick-and-place cell in a laboratory using calibration targets and physical test fixtures.

4: Validating before the shop floor
Physical trial-and-error during commissioning could increase integration effort and disrupt production. The solution therefore required high-fidelity simulation and iterative laboratory testing before shop-floor integration.

Solution

One coordinated intelligence layer from perception to execution

NTT DATA connected the virtual model, perception models, gripper strategy and cycle-control software into an end-to-end approach, then integrated and tested it in a laboratory environment.

Variant-ready pose estimation

Fine-tuned foundation pose-estimation models on target components, using a one-shot approach to accelerate adaptation to new variants.

Grasp strategy for complex parts

Extended the approach to more complex items, including the slider body and puller, and tuned the gripper strategy around their geometry.

Full-cycle control

Developed task execution and cycle management for the complete pick-and-place process, optimizing the control logic for a single-robot architecture.

Simulation-led validation

Built a high-fidelity simulation of the robot, gripper and parts with realistic visual and physical behavior, then integrated and validated the solution through iterative laboratory demonstrations and test-and-tune sessions.

Impact

A clearer path to industrial-scale picking

The initiative established an industrialization-ready foundation made up of reusable digital-twin assets, control software and a lab-validated technical roadmap. It supports lower-effort variant onboarding, fewer physical iterations before deployment and a simpler cell architecture. The source also identifies potential gains in picking stability, availability and scrap reduction; these outcomes remain subject to confirmation with production data and an agreed measurement baseline.

Drag