Demos that speak
louder than slides.
Six deployments across manufacturing, logistics, and assembly, captured on production hardware, not a demo rig.
Bin-picking with zero pre-sorting
Flexiv arm identifies, poses, and picks randomly oriented gear components from a mixed bin at production speed.
Real-time defect detection
Sub-millimetre anomaly detection on PCBs and machined parts. Flags defects before they reach assembly.
Guided assembly with pose feedback
6DoF pose loop guides the arm through multi-step assemblies with sub-mm placement accuracy.
Simultaneous tracking of objects
Real-time tracking of multiple parts on a moving conveyor, tracking ID, class, and 3D position maintained continuously.
Shared workspace collision avoidance
Detects human presence and intent in real time. Robot speed and trajectory adapt automatically within 8ms.
Pallet depalletising in unstructured stacks
Vision AI identifies package types, orientations, and stacking order in real time, with no fixed pallet format required.
Everything your robot needs
to see, decide, and act.
Six capabilities. One unit. Camera feed to grasp command, entirely onboard, no cloud round-trip.
Object Detection
Real-time multi-class detection at 60fps. Identifies and classifies every object in the workspace with bounding boxes and confidence scores.
Depth Perception
Sub-millimetre stereo depth maps in real time. Enables grasps and placements that work in production — not just in the lab.
6DoF Pose Estimation
Full 6-degree-of-freedom tracking of objects and end-effectors. Tells the robot not just where an object is — but exactly how it's oriented.
Edge Inference
All models run onboard. No cloud, no latency penalty, no data egress. Optimised for NVIDIA Jetson and Hailo. Works fully air-gapped.
ROS2 Integration
Native ROS2 nodes out of the box. Publishes object poses, depth clouds, and detection streams directly to your robot's topic graph.
Last-Mile Integration
Mounts on Flexiv, UR, Fanuc, KUKA and more. Full SDK. From unboxing to running inference on your arm: measured in hours, not months.
From camera to command
in four steps.
Most robots are blind. They follow fixed coordinates regardless of what's in front of them. We close the loop: camera feed in, pick command out, in under 5ms.
Perceive
Stereo cameras and structured light build a continuous 3D scene model — capturing geometry, depth, texture, and reflectance in real time. Not just a colour pattern. Full spatial understanding.
Understand
Object Intelligence classifies every item — known or unknown — without prior training. The system analyses an object's shape, orientation, and grasp affordances within milliseconds, adapting on the fly.
Decide
Edge inference computes grasp vectors, pick coordinates, and approach angles — all onboard, no cloud round-trip. Structured outputs ready for your robot's control system in under 5ms.
Executed.
Structured outputs feed directly into robot control systems via our low-latency API. Pick coordinates, grasp vectors, and object metadata, streamed at 60Hz, no middleware, no mapping layer.
The Humans
Behind the Robots.
Frequently Asked Questions
Everything you need to know about Yours, Physically — Spritle's Vision AI for industrial robots.
What makes Yours, Physically different? +
Yours, Physically by Spritle is an industrial robot Vision AI system designed for real-time robotic perception in industrial environments. Unlike cloud-based solutions, it runs fully onboard eliminating network latency and keeping data within your facility. It combines object detection, 6DoF pose estimation, and depth sensing into a single integrated system, enabling robots to operate in unstructured environments with minimal setup.
How can robots pick unsorted parts without trial-and-error? +
Yours, Physically functions as a bin picking vision system, enabling robots to understand mixed, unstructured scenes. It performs robot arm object detection and estimates object position and orientation in real time, allowing reliable grasping without pre-sorted inputs or fixed jigs. This improves efficiency and reduces manual intervention in industrial workflows.
Why does 6DoF pose estimation matter? +
6DoF pose estimation provides both the position and orientation of an object in 3D space. This allows robotic systems to determine the correct approach angle and grasp position, improving accuracy and reducing failed picks in real-world environments. It is a core component of any effective vision AI for robot arms.
How is Yours, Physically deployed? +
Yours, Physically supports ROS2 vision AI integration, publishing detection, pose, and depth data directly into existing robotic workflows. For non-ROS environments, a Python SDK (via gRPC) provides structured outputs for seamless integration across different systems.
Why run vision AI on-device instead of the cloud? +
Running onboard inference for robot arms eliminates dependency on internet connectivity and reduces latency. It also ensures that sensitive production data remains within the facility. This makes it more reliable for industrial environments requiring consistent, real-time performance.
Does it work with different robot arms? +
Yes. Yours, Physically is built as a vision AI for robot arms and supports major platforms including Flexiv, Universal Robots (UR), Fanuc, and KUKA. It provides structured outputs that integrate with standard robot control systems.
What kind of performance can be expected? +
The system delivers high-accuracy real-time object detection for robotics with low-latency inference. Performance may vary depending on hardware configuration and deployment conditions, but it is optimized for industrial-grade, real-time applications.
What production problems does it solve? +
Yours, Physically enables applications such as bin picking, sorting, assembly, and conveyor tracking using industrial robot vision AI. It reduces reliance on manual handling, pre-sorting, and fixed setups, enabling more flexible and scalable automation.
See your part get picked,
live.
We run the demo on a real robot arm. Bring a part number or just curiosity.
No deck. No SDRs. You talk directly with the engineers.
Typically responds within one business day
No spam, ever. We typically respond within 8hrs or less on business days.