www.ptreview.co.uk
19
'26
Written on Modified on
Teradyne Robotics showcases physical AI solutions at Automate 2026
Teradyne Robotics, the company behind Universal Robots (UR) and Mobile Industrial Robots (MiR), will demonstrate how physical AI is transforming industrial automation at Automate 2026 in Chicago.
www.universal-robots.com
Teradyne Robotics, the parent company of Universal Robots (UR) and Mobile Industrial Robots (MiR), will demonstrate how physical AI is transforming industrial automation at Automate 2026 in Chicago. The exhibition will take place from June 22 to 25, 2026, at booth #1250, showcasing deployable robotic applications designed for dynamic and unstructured environments.
Next-Generation Software and PLC-Style Logic
The foundational layer for these physical AI advancements is PolyScope X, UR’s next-generation software platform. While maintaining the core motion-control foundation of previous UR systems, PolyScope X modernizes the operator experience via a technology stack that includes modern web technologies, containerized applications, and native ROS 2 support.
A key feature of the software is the introduction of Logic Programs. These continuously running, multi-threaded programs execute in parallel with the main robot program. This native, PLC-style background logic allows programmers to coordinate and control multiple work cell components and exchange data independently of safeguard stops, program pauses, and the primary robot power state. Jean-Pierre Hathout, President of the Teradyne Robotics Group, noted that modern manufacturing demands an adaptable platform that integrates and evolves beyond the capabilities of a standalone robot arm.
Infrastructure Automation for Electronics and Data Centers
The exhibition highlights the company's strategic focus on infrastructure for electronics manufacturing and AI data centers through several technical showcases:
- The UR AI Trainer: Developed in collaboration with Scale AI, this imitation learning platform allows operators to physically guide a UR robot through assembly or packaging tasks. The system captures high-fidelity, force-aware data to train Vision-Language-Action (VLA) models for factory deployment. The platform supports cloud integration, demonstrating data transfers to train the NVIDIA GR00T open VLA model or utilize the NVIDIA Isaac Sim framework for validation.
- Generalist: Features two UR12e robots running autonomously on Generalist’s GEN-1 model. The setup showcases general-purpose robotic foundation models delivering physical world intelligence and dexterous manipulation at speed and reliability levels required for practical factory deployment.
- Cambrian: Developed to support the global build-out of AI infrastructure, this application uses dual-arm UR7e robots paired with Cambrian’s AI vision system to automatically identify and insert copper cables into high-density server racks.
Adaptive Material Handling and Ecosystem Integration
The exhibition includes a series of joint demonstrations with ecosystem partners, showcasing robots and autonomous mobile robots (AMRs) adapting to unstructured environments:
The exhibition includes a series of joint demonstrations with ecosystem partners, showcasing robots and autonomous mobile robots (AMRs) adapting to unstructured environments:
- AICA: Features a UR7e robot executing learned trajectories from human demonstrations. The robot utilizes force-sensing to pick up a metal part and buff it against a polishing wheel, matching the operator's speed and applied force based on a single demonstration.
- beRobox (PALTZ) + MiR Mobility: The PALTZ palletizer utilizes AI vision to redirect a UR20 robot on the fly, picking up boxes that have shifted orientation. The system operates alongside the MiR1200 Pallet Jack and a MiR600 AMR to retrieve pallets and coordinate material flow without fixed infrastructure.
- Mobile Cobot + ROEQ: Demonstrates a MiR250 AMR equipped with an ROEQ conveyor topper transferring components to a stationary conveyor line, where an MC250 mobile cobot manages return transport.
- Maple Advanced Robotics Inc.: An Autonomous Spot Sanding Solution utilizing a UR8L robot. The system requires no CAD models or manual path teaching; an operator marks defects, and the system scans and applies the correct finishing recipe automatically.
- Trener Robotics: Displays Acteris, an AI-native platform with a conversational interface. Operators can deploy robotic machine tending jobs via simple chat input in any language, setting up new tasks for a UR7e robot within minutes.
- Vention: The Rapid Operator AI bin-picking solution utilizes a UR12e robot and 3D vision to handle unstructured parts for real-time identification, achieving a high first-pick success rate.
Additional Context
This section details technical specifications not included in the original news release.
Implementing physical AI in industrial robotics requires transitioning from deterministic, pre-programmed path trajectories to real-time, sensor-driven reactive control. Traditional industrial robots operate on absolute joint coordinates, executing identical kinematic paths repeatedly. If a target object shifts by a few millimeters or its orientation changes, the robot fails to grasp it, often triggering a mechanical collision fault. Physical AI models, such as Vision-Language-Action (VLA) architectures, bypass fixed coordinate programming by mapping high-dimensional sensory inputs—like 3D point clouds from RGB-D cameras and multi-axis force/torque data—directly to low-level motor joint velocities.
This end-to-end control is optimized through imitation learning and foundation models like GEN-1. During the training phase, a human operator guides the manipulator through a task using a haptic feedback device or direct kinesthetic teaching. The robot's internal encoders log exact joint positions while specialized strain-gauge sensors measure torque vectors at the tool flange.
The resulting dataset couples spatial trajectory coordinates with applied physical forces. Neural networks process this multimodal data to build a generalized execution policy. When deployed on a state-of-the-art tech stack like PolyScope X with native ROS 2 (Robot Operating System) communication nodes, the robot can compute dynamic path corrections mid-trajectory. This edge-computed processing enables the arm to modulate its clamping force and approach angles on the fly, mimicking human dexterity when executing surface finishing, cable routing, or bin-picking operations in unstructured environments.
Edited by Romila DSilva, Induportals Editor, with AI assistance.
This section details technical specifications not included in the original news release.
Implementing physical AI in industrial robotics requires transitioning from deterministic, pre-programmed path trajectories to real-time, sensor-driven reactive control. Traditional industrial robots operate on absolute joint coordinates, executing identical kinematic paths repeatedly. If a target object shifts by a few millimeters or its orientation changes, the robot fails to grasp it, often triggering a mechanical collision fault. Physical AI models, such as Vision-Language-Action (VLA) architectures, bypass fixed coordinate programming by mapping high-dimensional sensory inputs—like 3D point clouds from RGB-D cameras and multi-axis force/torque data—directly to low-level motor joint velocities.
This end-to-end control is optimized through imitation learning and foundation models like GEN-1. During the training phase, a human operator guides the manipulator through a task using a haptic feedback device or direct kinesthetic teaching. The robot's internal encoders log exact joint positions while specialized strain-gauge sensors measure torque vectors at the tool flange.
The resulting dataset couples spatial trajectory coordinates with applied physical forces. Neural networks process this multimodal data to build a generalized execution policy. When deployed on a state-of-the-art tech stack like PolyScope X with native ROS 2 (Robot Operating System) communication nodes, the robot can compute dynamic path corrections mid-trajectory. This edge-computed processing enables the arm to modulate its clamping force and approach angles on the fly, mimicking human dexterity when executing surface finishing, cable routing, or bin-picking operations in unstructured environments.
Edited by Romila DSilva, Induportals Editor, with AI assistance.
