Assembled · horizontal
MARU Link
Four sensor modules — vision wide, vision narrow, ToF 4-VCSEL, ToF 1-VCSEL — dock into three edge bodies over Ethernet, USB-C, or SerDes. Same connector on every module, so a change of configuration is a swap.
Who we are
Edge AI Vision Infrastructure for Physical AI
We bridge the perception gap for robotics and automation: 3D Edge AI perception modules already in production — 4,000+ units shipped.
01 / 02
Problem
Physical AI Starts with Accurate Perception
READY
Brain — AI
Foundation models keep getting more capable and cheaper — every year.
MISSING
Perception
Seeing and understanding the real world precisely. Everything else depends on it.
READY
Body — robot hardware
Actuators and chassis are commoditizing fast as volumes scale.
“Fine components cannot make a fine result.”
WHAT MAKES IT HARD
Any one of these is tractable on its own. Holding all four at the same time, outside the lab and across a full shift, is the part that has stayed unsolved.
Depth accuracy
A distance is only useful if the machine can act on it. Error that reads as negligible on a datasheet becomes a missed grip or a late stop at the point of contact.
Thermal drift
A module calibrated on the bench warms up in service, and its optics and sensor move with the heat. Calibration that was true at power-on stops being true, and the data gives no warning.
Sensor registration
Depth and color are assumed to describe the same point in space. Mounted and timed separately, they do not — and every measurement taken downstream inherits the offset.
Deterministic latency
An answer that arrives after the control loop has moved on is not an answer. Perception has to finish inside the machine’s cycle, on the machine, every cycle.
“$3–5 of integration for every $1 of hardware.”
WHY IT STAYS HARD
That integration cost is structural. The pieces a working perception system needs are each developed at a different company, and nobody owns the seams between them.
Imaging module
Sensor and optics, across a range of different image sensors.
Camera firmware
For module bring-up and setup.
Vision controller
For the AI vision solution.
Machine vision SDKs
Various, and AI-based.
“Who handles the integration?”
Solution
Modular Perception, Sensors to On-Device AI
We do. The two halves of the platform ship together, and there are two ways to put them to work.
Precision sensing
Depth and fused imaging at the source. Metric depth for every pixel, from a sensor priced for volume.
Edge intelligence
Detection and pose computed on-device. Server-grade spatial AI, running on the robot.
Industrial scale
Same blocks, recombined per customer. Application-specific configuration, without rebuild.
“It arrives calibrated, it holds steady outdoors, it updates itself in the field, and it drops onto whatever compute they already chose.”
Two ways to work with us
Which one fits is settled by profiling your environment, not by picking from a page.
MARU module supply
Your environment maps onto blocks that already exist. We fix the module mix and tune it against your profile before hardware ships. What you buy is a configured product.
How this worksCo-engineering
Your environment falls outside what a configuration can reach. We design with you on the same stack, so the result stays reusable instead of becoming a one-off.
How this worksProduct lineup
Three ways to build the same platform
MARU Gen 2 assembles horizontally, stacks vertically, or arrives integrated. The perception blocks are the same in all three.
Assembled · vertical
MARU Tower
Sensor, edge, and interface stages stacked on one vertical axis — the smallest installed area of the three builds, for cells and cabinets where bracket space is the constraint.
All-in-one
MARU ONE
Optics, edge compute, and I/O in one sealed housing. One mount, one cable, nothing to assemble on the line — for volume deployments where install time dominates.
Key features
What every build does
2D & 3D sensing
ToF · VCSEL 940 nm · 0.2–20 m · ±0.5%
RGB + IR · 5 MP–8K
On-device AI
Detection · segmentation · 6-DoF pose
Runs on the edge SoC, no cloud
Deterministic latency
10–20 ms on-device
Inside the machine cycle
Industrial interfaces
USB-C · Ethernet · SerDes · HDMI/DP
ROS 2 · REST/SDK API
Applications
Where MARU is deployed
Robot makers · System integrators · Enterprises in manufacturing, logistics, and healthcare.
Robotics
Arm guidance, bin picking, mobile robots, and service robots.
Smart Factory
In-line inspection and cell-level safety.
Agriculture
Crop and row perception for field robots.
Logistics
Pallet and parcel handling, in-line volume and dimension measurement.
Healthcare
3D vision for medical-device equipment.
Traction
In production, not in pilot
The figures below are perception and camera modules — shipped, in the field, today. MARU Gen 2 is built on that production base and reaches the market in March 2027 — evaluation kits of the same stack are available now.
4,000+
Perception and camera modules shipped to date
10+
Paying customers
In production
Perception and camera modules mass-produced on our own lines
16
People — incl. 10 senior R&D engineers, 15+ yrs avg
“C-suite from the Namuga · Samsung · Qualcomm ecosystem.”
Ecosystem
Companies in the MARU supply and deployment chain
