Edge Intelligence in Action
The robotic arm demo, built for Macnica, connects on-device vision, natural-language understanding, and robotic action in one local Edge AI pipeline.
Get in touch
How the System Works
Fully on-device Edge AI stack
The demo combines two local AI workloads with a robotic system: vision understands the scene, GenAI interprets the instruction, and the robot turns the result into physical action.
The visitors can enter free-form commands or use predefined prompts; the Ambarella N1-655 runs the language model locally, while the iENSO Mako camera with Ambarella CV75m runs the vision models and provides scene awareness. The UR7e cobot then performs the sorting action.
Prompt
The visitor enters a natural-language instruction, such as which shapes or colors should be moved and where they should go. The Ambarella N1 processes the instruction locally and turns it into structured commands.
Perceive
The camera identifies the objects on the worktable and provides the visual context needed to match the instruction to the physical scene. An iENSO Mako camera with an Ambarella CV75m runs the vision models on-device.
Act
The system combines the instruction with the visual result and sends the required action to the robotic arm, which picks and sorts the selected objects. The UR7e cobot executes the movement using a custom electromagnetic end-effector.
Visual recognition, and prompt processing can happen locally.
The demo runs vision and language models directly on Edge AI hardware. The system interprets the scene and the user instruction locally, reducing dependence on cloud infrastructure and keeping the decision loop close to the device.
Vision and AI become more useful when their output drives what the system actually does.
Our demo connects visual understanding and natural-language intent to robotic motion. The result is a complete perception-to-action loop in which AI output becomes an actual physical response.
The difficult part is making every layer behave like one system.
Camera input, vision models, language processing, application logic, robotic control, and the user interface all have to work together continuously. That integration challenge is typical of real embedded vision and Edge AI products.
Why Teams Build With Us
They need to make it work on the Edge
The AI works in development, but the target device changes the equation.
Too slow, large, or inaccurate
Models that perform well on a GPU may be too slow, too large, or lose accuracy once moved to an embedded SoC. Runtime, memory, sensor pipelines, and the surrounding firmware stack all start to matter.
Edge expertise
We port, quantize, and optimize the workload for the target platform, then integrate it into the wider embedded vision system — from board support and sensor input through inference to application logic.
They need to get there in time
The technical direction is clear, but the internal team cannot deliver fast enough.
Deadlines won’t move
A trade show, pilot, product milestone, or investor demo is approaching while the existing engineering team is already stretched. Recruiting and ramping up specialist embedded vision engineers can take months.
Specialized capabilities
We are an experienced engineering team that can take ownership of a defined part of the roadmap and start contributing without waiting for a long hiring cycle.
They need to get it market ready
The prototype works, but the product still has a long way to go.
Production blockers
Platform fit, performance, integration, certification, security, or maintainability can become blockers once a prototype has to operate reliably outside controlled conditions.
Deployment support
We help close the gap from prototype to deployable product — from hardware and architecture decisions through optimization, system integration, certification support, and field readiness.
What We Get Involved For
Get in Touch
Looking for an Edge AI technology partner?