JTEKT Physical AI Trial Achieves 97.1% CVJ Picking Success

JTEKT announced on November 2, 2026, that its physical AI trial on the constant velocity joint (CVJ) production line at the Tadomisaki Plant achieved a 97.1% success rate in random picking. The process involves identifying and picking up parts positioned randomly rather than in predetermined locations. Under challenging test conditions, the system improved the success rate by 12.1 percentage points compared with conventional control, which recorded 85.0%. It also reduced the average cycle time by 16%, demonstrating improvements in both picking accuracy and operational efficiency.

The trial was conducted jointly with Upstage AI and KARAKURI between March 19 and August 31, 2026, under AWS Japan's physical AI development support program. The participating companies deployed AI trained in a virtual environment onto a robot for operation on the production line. Using camera images, the system recognizes the positions and conditions of individual parts before autonomously determining appropriate grasping positions and transfer motions. This approach enables the robot to respond to randomly positioned components without relying entirely on predefined picking locations.

The system combines high-speed position detection AI with high-precision shape recognition AI. The integration is designed to deliver both accurate recognition and rapid processing, addressing two important requirements for automated random-picking operations. Position detection identifies where parts are located, while shape recognition supports the assessment needed to determine suitable grasping positions. By combining these capabilities, the system achieved the reported 97.1% picking success rate under the trial conditions while also reducing average cycle time compared with conventional control.

The trial also demonstrated that productivity improvements could be achieved without major modifications to existing production equipment. Adding an AI PC to the existing equipment was sufficient to introduce the physical AI capability, offering an approach that can build on established manufacturing infrastructure. This finding is relevant to the implementation of AI in existing production environments, where substantial equipment changes may otherwise be required. JTEKT intends to use physical AI to improve success rates and eliminate human intervention in processes that are already automated, while also supporting the transfer of skills in manufacturing operations that depend on experienced workers.

Industry Impact & Outlook

JTEKT's trial highlights the potential of physical AI to improve manufacturing automation through more accurate robotic handling and shorter production cycle times. The combination of virtual-environment AI training, camera-based recognition and autonomous motion planning demonstrates an approach to handling randomly positioned components on existing production lines. The ability to introduce the system by adding an AI PC, without major equipment modifications, also provides a practical route for upgrading established manufacturing operations. JTEKT's plans to apply physical AI to both automated processes and operations dependent on experienced workers indicate its intended role in improving productivity while supporting the transfer of manufacturing skills.

Frequently Asked Questions

What success rate did JTEKT achieve in its physical AI trial?
JTEKT achieved a 97.1% success rate in random picking during its physical AI trial on the CVJ production line. The trial took place at the company's Tadomisaki Plant under challenging test conditions. This represented an improvement of 12.1 percentage points compared with conventional control, which recorded an 85.0% success rate. The system also reduced average cycle time by 16%, demonstrating improvements in both picking performance and processing efficiency.

Where was JTEKT's physical AI trial conducted?
The physical AI trial was conducted on the constant velocity joint production line at JTEKT's Tadomisaki Plant. The project focused on random picking, which requires robots to identify and handle parts positioned unpredictably. JTEKT worked with Upstage AI and KARAKURI to develop and test the system between March 19 and August 31, 2026. The trial was conducted as part of AWS Japan's physical AI development support program.

How does JTEKT's physical AI system perform random picking?
JTEKT's physical AI system uses camera images to recognize the positions and conditions of randomly positioned parts before determining how to pick and transfer them. AI trained in a virtual environment is deployed onto a robot, allowing it to autonomously determine suitable grasping positions and transfer motions. The system combines high-speed position detection AI with high-precision shape recognition AI. This combination supports both recognition accuracy and processing speed, enabling more reliable random-picking operations.

How much did the physical AI system improve production performance?
The physical AI system increased random-picking success from 85.0% under conventional control to 97.1%, an improvement of 12.1 percentage points. It also reduced average cycle time by 16% during the trial. These results indicate improvements in picking reliability and operational speed under the tested conditions. JTEKT reported the performance outcomes from its CVJ production line trial at the Tadomisaki Plant.

Does implementing JTEKT's physical AI system require major equipment modifications?
No. The trial confirmed that productivity improvements could be achieved without major modifications to the existing production equipment. The system was introduced by adding an AI PC to the existing equipment. This approach allows physical AI capabilities to be incorporated into an established manufacturing setup without requiring substantial equipment changes. The finding demonstrates the implementation approach tested by JTEKT during its CVJ production line trial.

What are JTEKT's future plans for physical AI?
JTEKT plans to use physical AI to improve success rates and eliminate the need for human intervention in processes that are already automated. The company also intends to apply the technology to support the transfer of skills in manufacturing operations that depend on experienced workers. These objectives extend the intended application of physical AI beyond improving robotic picking performance, covering both existing automated processes and operations where experienced workers' skills remain important.