Giving machines a human-like sense of sight.
Datellx's mission is to give robots human-like senses, starting with vision. Our first product, Instinct, is a camera-based perception system: point it at a scene and it works out what's there, what it's made of, and how it relates to everything around it, including objects it has never seen before.
Even the best-funded robots still struggle with the object they've never seen before.
Real environments like homes, warehouses, and resale streams are full of objects that don't fit any predefined list. Most vision systems simply stop working when they hit something unexpected, instead of reasoning their way through it.
Limited to what it already knows
Most vision systems are built to recognize a fixed list of categories. The moment a scene includes something outside that list, they stall, or guess.
No sense of what it's made of
Most systems can tell you where an object is without telling you what it's made of: glass, fabric, foam, or metal, the details that matter for handling and value.
Objects seen in isolation
Most systems can point to individual objects but rarely reason about how they relate: what's stacked on what, what's inside what, what's supporting what.
A perception layer built to reason about the unknown.
Instinct is designed to take in a scene through any camera and work out what's in it, including objects it has never encountered before, by connecting the unfamiliar to what it already knows rather than stopping at a fixed list of categories. It's built to work with the equipment a team already has, not to require anything specialized.
Reasoning about the unknown
Built to connect objects it has never seen before to what it already knows, instead of stalling on anything unfamiliar.
What's there
Identifies and separates every object in a scene using nothing more than a standard camera.
What it's made of
Estimates material and condition (glass, fabric, plastic, metal, wear and damage), the properties that matter for grading and handling.
How it relates spatially
Reasons about how objects sit relative to one another, so a scene reads as a whole, not a list of disconnected items.
Instinct focuses on understanding a scene. It's designed to work alongside the systems that plan and act, not replace them.
Built for teams that need to understand unfamiliar objects.
Instinct is a perception layer other systems build on top of.
Robotics & physical AI
Teams that would rather buy a perception layer than build one in-house, so they can spend their effort on planning and control.
Resale & liquidation
B2B resale and liquidation platforms that need consistent condition and material grading across a stream of hard goods they've never catalogued before.
Home & insurance
Home-contents and insurance platforms that need to inventory and value the unfamiliar mix of items found in a real home.
Vision first. Other senses next.
Long term, our goal is to give robots human-like senses: not just cameras that react to what they see, but the kind of layered understanding a person builds from sight, touch, and everything else they perceive. We're starting with vision because it's where the gap between what robots can sense and what they can actually understand is most obvious.
Building Instinct: visual perception and reasoning about unfamiliar objects, from any camera.
Extending that same reasoning toward other physical senses, as the vision layer matures.
Building Instinct.
Want an early look, or think Instinct could fit what you're building? Reach out. We read every message.
hello@datellx.comDatellx LLC · Madison, WI