Harlond
AI, off the screen
and onto the floor.
Physical AI starts here.
Harlond builds physical AI: robots that see, learn and move on their own. We take intelligence trained in simulation onto the real floor, so any factory can put intelligent robots to work.
- Vision AI
- Sim-to-real
- AI services
- PoC & joint R&D
- Manipulators · Cobots · AMRs
The problem
Why robots are still hard to teach
Most robots on a factory floor are brilliant at one fixed motion and helpless the moment something changes — a new part, a different bin, a cloudy afternoon through the skylight.
- Slow
Teaching takes specialists
Vision setups, grasp points and paths are still tuned by hand, one cell at a time. Smaller manufacturers rarely have those specialists in-house.
- Risky
Practice on the line costs
Trial and error next to production costs time, parts and safety margin. Simulation is cheaper — but what works in sim often stumbles for real.
- Fiddly
Plugging in is a project
PLCs, production systems and monitoring all need wiring together — and every change on the line means doing some of it again.
Our approach
Trained in simulation. Proven on the floor.
- Simulated twin
- Practice at scale
- Field validation
- Real floor
- Learn & improve
We rebuild a robot’s working world — the arm, the parts, the cameras, the fixtures — as a physics simulation grounded in real measurements. There, perception and control policies can practise thousands of variations of light, pose and motion, and every mistake is free.
Then we carry them over to the real floor: checking against real data, closing the gap with calibration and careful fine-tuning, and keeping hard safety limits around everything a learned policy does. Once a skill is at work beside your PLC, what it sees flows back to make the next version better.
Around all of this sit AI tools meant for the people on the line, not just engineers: describe a task in plain words, ask why a cell stopped, or get help mapping signals to a PLC — so adopting robot intelligence feels less like a research project and more like a conversation.
What we build
Three tracks
We develop technology and services hands-on with the people who’ll use them — through proofs of concept and joint R&D with manufacturers, integrators and labs.
-
Vision AI
Perception that copes with real conditions: jumbled bins, reflective metal, changing light. 2D and 3D, from finding a part to checking it’s right.
- Bin picking
- Pose estimation
- Visual inspection
- Calibration
-
Sim-to-real learning
Reinforcement- and machine-learning policies trained in simulation and carried over to real hardware — for grasping, placing, insertion and navigation.
- Manipulators
- Cobots
- AMRs
-
AI services
Agents and tools that make setup, teaching, monitoring and integration simpler for the people who run the line every day.
- Natural-language setup
- Line monitoring
- PLC integration
Questions we’re working on
- Sim-to-real transfer — policies that hold up on hardware with little real-world tuning
- Domain randomization — varied visuals and physics so models generalize
- Vision-language-action — linking what a robot sees, what you ask and what it does
- Safe collaboration — learned behavior that respects force, speed and shared space
- Data engines — synthetic and field data in one steady improvement loop
Simulator solution · Coming soon
From the data you need to a working cell
Harlond Simulator guides you step by step: what data your task needs, how to refine it, what it can do, how to test and tune it virtually — and how to bring it to the real line.
Define
The data you need
Start from the task. Harlond Simulator works out which data it needs — camera views, part poses, robot and PLC signals.
Refine
Collect, clean, label
A clear plan for how to collect, clean and label that data, so it’s ready for training and testing.
Discover
What it can do
See which vision and automation tasks your data can support — before you commit to any of them.
Test & tune
Virtually, first
Run the task on that data in simulation and tune it there, where mistakes cost nothing.
Deploy
On the real line
Bring it to the floor with sim-to-real validation, with PLC signals and interlocks taken into account.
Robots we work with
One way of learning. Many kinds of robot.
Manipulators
Fixed industrial arms that pick, place, tend machines and assemble.
Cobots
Multi-joint collaborative robots that share a workspace with people.
AMRs
Mobile robots that carry parts between cells and around a busy floor.
How to work with us
Bring us your “too hard” list
Joint R&D
Explore an open question together — sim-to-real, perception or AI-assisted operation.
Proof of concept
Try an approach on one real task in one real cell before committing further.
Service development
Shape AI tools for setup, monitoring or integration around how your team actually works.
We’d especially like to hear from small and mid-sized manufacturers, robot system integrators and research labs connecting sim-to-real work with real hardware.
Contact
Come say hello
Tell us about your robot, your cell or your research question.
PoC or joint R&D — it all starts with an email.