Recognize
Use infrared images as input to suitable recognition software on your host. The model and its training data determine which objects it can identify. Check that it supports your image input.
Physical AI brings sensing, software and action together in devices that interact with their surroundings. A robot locating an object, a touchless control responding to a hand and an appliance checking placement are examples.
Your first experiment can fit on a desk. Measure the distance to a mug, then make an on-screen indicator respond. Start with sensing data and a simple rule in your software. Build from what you can observe.
A suitable recognition model can use an image to identify an object. Recognizing a mug tells your application what it is looking at. It does not supply a measured distance.
Depth adds distance and position information. Your software can associate those measurements with an image region to explore where a recognized object sits.
Consider a possible mug-finding prototype. A suitable model on your connected computer could identify the mug in an infrared image. Your application could associate available depth points with that region and display a label and estimated distance.
That is a proposed application workflow, not a ready-made Pico feature. Start with an on-screen result. Moving a robot toward the mug requires planning, control and further testing.
Use infrared images as input to suitable recognition software on your host. The model and its training data determine which objects it can identify. Check that it supports your image input.
Add measured distance to your application’s view of the scene. Check the available depth coverage and how your software associates image regions with depth points.
Let your application decide what happens next. Display a distance, change an indicator or send a command to another device. Test the response in your intended scene.
Pico is a compact USB 3D depth camera for makers. It supplies point clouds and infrared images for experiments with distance, object location and interaction. You bring the application software and the idea.
Pico’s Raspberry Pi RP2354 microcontroller processes depth on the sensor. Your host is the supported computer or Android device connected by USB. It supplies USB power, receives Pico’s data and runs MkE Connect or your application software.
Use MkE Connect for Windows, macOS and Linux, plus Android 8.0 or later with USB OTG. Check the release compatibility guide for exact requirements. Recognition models need their own compatible host software and processing resources.
A separate Raspberry Pi computer is optional. Pico already contains a Raspberry Pi microcontroller for depth processing. That built-in chip does not establish compatibility with every Raspberry Pi computer. Confirm support for your intended host.
Pico supplies sensing data. Object recognition and AI-driven decisions require suitable host software. The current evaluation configuration provides infrared images, not RGB color photographs. Combined mode alternates depth and image captures, so account for motion between frames.
These are application concepts, not demonstrated Pico integrations. Each needs suitable host software, integration and testing. Begin with a visible result on screen before adding physical movement.
Put a stationary object in view and explore its depth points in MkE Connect. Move it closer, then farther away. Watch the data change and learn which surfaces return measurements.
Next, write a rule that changes an on-screen indicator as measured points enter a chosen distance zone. Add object recognition after checking the model’s infrared input requirements. Keep each step small enough to test.
Give your next idea a practical first step. Get Pico launch updates for availability and setup news, or request early access to purchase pre-release hardware for your experiments.