Overview
From December 3–6, 2025, Eureka Robotics exhibited at the International Robot Exhibition 2025 (iREX 2025) held at Tokyo Big Sight. Under the theme “Physical AI that Delivers”, we showcased seven live demonstrations highlighting how our Physical AI technologies solve some of the most challenging automation tasks.
Below are the highlight videos from each demo we presented at the show.
We invite you to share them widely within your teams and departments.
Random 3D Picking & Classification
In pattern-projection 3D Vision systems, glossy surfaces often lead to unstable recognition. When automating multiple steps like bin picking, front/back identification, and alignment, the integration load increases, raising both cost and complexity.
Eureka Robotics Solution
✔ Stable glossy-part recognition: Avoids challenges from pattern projection based systems
✔ Easy system integration: One Eureka Controller for Robot, 3D and 2D cameras, and other devices
✔ All-in-one development environment: Drivers, AI studio, simulator, and intuitive tools included
CAD-based Visual Inspection
Manual comparison between CAD data and the actual part is highly labor-intensive. For large workpieces, inspection requires multiple cameras and dedicated fixtures, limiting layout flexibility. Robot-based inspection also demands teaching numerous inspection points, driving up integration effort.
Eureka Robotics Solution
✔ Instant inspection with CAD × AI: A single camera instantly checks hole positions, screw presence, and more using AI.
✔ No precise positioning required: No need for strict camera/part alignment or robot integration
✔ Covers large parts with one camera: Simply capture the required areas—AI maps them to the CAD model.
Random 3D Picking & Insertion of Connectors
Mating requires high insertion force and precise “tactile” sensitivity to ensure a reliable fit.When picking from a moving conveyor without stopping it, the robot must also perform highly accurate motion tracking.
Eureka Robotics Solution
✔ Eureka 3D camera and AI enable precise picking of moving workpieces
✔ With force-control technology, the robot senses contact between the connector and socket, and inserts reliably until full engagement
✔ A single Eureka Controller manages both vision and force control
✔ Fast insertion enabled by a high-speed force-feedback control loop
High Speed, Random 3D Bin Picking
High-precision 3D point cloud generation traditionally requires significant processing time, making it unsuitable for high-speed operation.In bin-picking, the wide variation in part orientations requires AI training rather than manual registration, but hand-annotating training data demands substantial effort and time.
Eureka Robotics Solution
✔ AI-based fast 3D reconstruction: full vision pipeline < 0.5 sec
✔ Sim2Real: vision models trained with purely synthetic data. No manual annotation, deployable in a day. Handles reflective items, ensures consistent orientation picking
✔ Single interface for control: the Eureka Controller
Masterless Random 3D Picking
In high-mix production, where many different parts must be handled, creating a universal AI model, or individual models for each part, is practically impossible.
Eureka Robotics Solution
✔ Masterless: No CAD, no prior data registration
✔ Super Model: One model handles everything from item recognition to grasp estimation
✔ High-Speed 3D Vision: Capture to grasp output in under 1.5 seconds
Masterless Random 3D Picking
Parts with narrow, intricate geometries or glossy metallic surfaces often produce unstable 3D point clouds.In bin-picking, part orientation changes every time, making it difficult for traditional rule-based methods to extract reliable features—one of the key barriers to automation.
Eureka Robotics Solution
✔ Robust 3D perception: Consistently generates high-quality point clouds on parts with diverse materials and surface properties (metallic, plastic, reflective).
Stable even during in-hand movement
✔ Intelligent part understanding: AI models recognize and pick complex, intricate geometries with high reliability
✔ Precise motion execution: Advanced motion planning and control enable accurate pick-and-place across varying orientations
マスターレスピッキング 体験コーナー
We demonstrated how simply placing a part triggers instant 3D data generation, after which AI automatically estimates the grasp position and orientation, allowing the robot to pick it immediately.
Eureka Robotics Solution
✔ Simplified Interface: from setup to picking
✔ Masterless: No CAD, no prior data registration
✔ Super Model: One model handles everything from item recognition to grasp estimation
✔ High-Speed 3D Vision: Capture to grasp output in under 1.5 seconds
Contact Us Anytime!
If you are exploring automation but are unsure where to start, or if you have tried in the past without success, we are here to help.
At Eureka Robotics, we tackle challenges that conventional 3D vision systems have been unable to solve and deliver practical, reliable solutions.
If you are facing automation difficulties, please feel free to reach out to us anytime.