We help you solve real-world use‑cases using robots
Ready-trained vision models that drop straight into your robots: pick the use cases, load the models, and the robot sees and acts. No advanced training required.
The hardware already moves. Ripik supplies the eyes and the judgement, so the robot can act. Proven across 300+ deployments in 150+ plants, in iron & steel, metals, cement, mining, energy and pharma.
Move + See + ActIndustry example: robotic hot-metal work
Move + See + ActIndustry example: robot that sees and acts
Move + See + ActIndustry example: gallery patrol
The offer · Three levels of a robot
A robot can move, see, or see and act
Most robots on plant floors today stop at the first two levels. Level 3 is where Ripik comes in.
Industry example: fixed arm cell
1Move
A machine that moves exactly as it was programmed or driven. It repeats a path or follows a joystick. It does not know what is in front of it.
Where most robots are today
Industry example: quadruped patrol
2Move + See
The robot carries cameras, takes in the scene and makes sense of it: what is this object, is it out of place, is it hot, is it moving. Ripik already does this for plants with computer vision.
Ripik does this today
Industry example: robotic hot-metal work
3Move + See + Act
The robot understands what it sees and does something about it: stops the belt, raises the alarm, clears the rock, re-routes the truck. Vision becomes a decision and the decision becomes an action.
What we have to offer
So whatThe hardware already moves. Ripik supplies the eyes and the judgement, so the robot can act. Level 3 is our expertise.
The offer · What you get
Your robot + a Ripik model = a robot that acts
Pre-trained vision models, packaged to run on the robot you already have or are about to buy.
Industry example: legged robot
Your robot
Legged, wheeled, rail, drone or arm. Any carrier with a camera.
+
Big Rock DetectionFurnace MonitoringFlare StackCoil GradingOSHA ComplianceForeign Object Detection
Ripik vision models
42 use cases across 6 industries. Load one or load them all.
=
Industry example: robot that sees and acts
A robot that sees and acts
Detects, decides and triggers the action, on site, from the first shift.
How it plugs in
1
Camera feed in
The robot's own optical or thermal camera. A standard video stream (RTSP, USB, GigE) is all the model needs.
2
Ripik model on the edge
Runs on the robot's compute or an edge box beside it. Detects and decides on site, with no cloud round-trip.
3
Event out to your systems
Detection, decision, location and frame go to the robot controller, PLC, SCADA, MES or an alert app.
So whatAny robot that exposes its camera stream and accepts an event can carry Ripik models. Connects to CCTV, SCADA, MES and SAP today.
The offer · Proof
Already proven across 300+ deployments worldwide
The same models run today on fixed cameras across iron & steel, metals, cement, mining, energy and pharma.
300+
AI deployments
across 150+ plants in iron & steel, metals, cement, mining, energy and pharma
50+
enterprise customers
across 15+ countries
$20M+
annual savings
delivered to clients
95%+
model accuracy
on live plant footage
Published results:95% accuracy on refractory monitoring · 70% less conveyor-failure downtime · 70% fewer switchyard failures · 93%+ coal-moisture accuracy
Ripik.ai is trusted by
Alerts are triggered when particles exceed the set threshold, and it works very well.
Digital Transformation Technology Lead
Holcim
Positively impressed about the number of strange objects that were detected on the belts.
Chief Digital Officer
ArcelorMittal
The weight of material measured in real-time seems very close to what we estimated there being.
Melt Shop Supervisor
Nucor
What used to be static visuals on the wall have transformed into real-time, actionable information.
Senior Process Engineer
Martin Marietta
So whatThe vision is already trusted on fixed cameras at the biggest names in heavy industry; a robot is simply a new carrier for the same models.
The offer · Benchmark
Benchmarked against five public models
400,000 plant images. Recall is the share of real events caught; precision is the share of alerts that were real.
Recall / precisionreal events in 100,000 images
Gemini 3.1 ProFrontier VLM
Qwen3-VL-4BLocal VLM
Florence-2Open detector
Grounding DINOOpen detector
YOLO-WorldOpen detector
RipikProduction model
Big Rock Detection48,981 real events
0.01 / 0.78
0.04 / 0.70
1.00 / 0.49
1.00 / 0.49
0.46 / 0.69
0.98 / 0.97
Foreign Object Detection14,246 real events
0.00 / 0.00
0.05 / 0.23
0.15 / 0.20
0.07 / 0.46
0.00 / 0.00
0.93 / 0.97
Conveyor scratch detection52,720 real events
0.20 / 0.66
0.04 / 0.58
0.06 / 0.57
0.56 / 0.51
0.33 / 0.61
0.93 / 0.98
Thermal hotspot detection60,067 real events
0.72 / 0.66
0.63 / 0.81
0.08 / 0.64
1.00 / 0.60
0.70 / 0.56
1.00 / 1.00
Speedseconds per image, lower is faster
3.70 scloud only
0.38 son the edge
0.87 son the edge
0.65 son the edge
0.14 son the edge
0.40 son the edge
Usable: both scores 0.70 or abovePartial: weakest score 0.30 to 0.69Unusable: weakest score below 0.30Florence-2 and Grounding DINO reach 1.00 recall by flagging every image, so precision collapses.
So whatGeneral-purpose vision models have never seen a belt, a ladle or a kiln. Ripik's have, across 300+ deployments, and the scores show it.
What the scores mean on a shift · per product, 100,000 images
So whatA robot carrying a general-purpose model walks past most of the problems. A robot carrying Ripik's catches them, with almost no false alarms.
Use cases · The model catalogue
42 ready models for material, equipment and safety monitoring
Every model here already runs on fixed cameras in plants. Put the same model on a robot and it goes where a fixed camera cannot.
Best suited to a moving camera: patrol, drone or vehicle-mounted.
Material monitoring on a robot · 19 models: sizing, moisture, stockpiles and foreign objects coming in; coils, slabs and sheets going out
Industry example: inspection drone
Stockpile Volume Estimation
Drone surveys the yard
A drone or rover carries the stockpile model over the coal and limestone yards each shift. The robot plans the pass, the model returns volume and shape, and the inventory updates without a survey crew.
Industry example: coil-handling robot
Coil Grading + Coil Tracking
Robot grades every coil
An arm or rail robot at the coil store carries the grading and tracking models. Each coil is identified, surface-graded and logged as it is placed, and a reject is flagged before it ships.
Industry example: quadruped patrol
Big Rock + Foreign Object
Patrol watches the feed
Big Rock and Foreign Object models ride a patrol along the feed belts ahead of the crusher and the mill, so an oversize lump or a metal strap is caught before it reaches the liner.
Equipment monitoring on a robot · 17 models: conveyors, ladles, furnaces and switchyards; burners, flares, kilns and sinter beds
Industry example: gallery patrol
Belt + Sway
Belt gallery patrol
Belt and sway models ride the patrol; a tear or a drift raises the alarm mid-round.
Industry example: in-kiln drone
Kiln Health
Drone in the kiln
On shutdown a caged drone flies the kiln model along the shell and maps wear.
Industry example: quadruped patrol
Furnace + Switch Yard + Noise
One patrol, several models at once
A walking robot carrying furnace, switchyard and noise models does the shift round once: it checks refractory hot spots, scans transformer and rectifier temperatures and listens for an abnormal bearing, and raises each as a separate event with the frame attached.
So whatA patrol robot carrying ten of these models replaces ten fixed-camera installs and still reaches the places no camera is mounted.
Safety monitoring on a robot · 6 models: people, vehicles, procedures and recipes, checked on every round
Industry example: quadruped patrol
OSHA Compliance
Quadruped checks PPE
The OSHA model on a walking robot spots a missing vest, helmet or harness while it patrols, and logs the frame so the shift lead can act before the next round.
Industry example: patrol on plant roads
Road Safety + Truck Trip
Watches the haul roads
Road safety and truck trip models on a robot or roadside unit flag speeding, wrong-lane movement and unsafe loading on plant roads where no camera is mounted.
Industry example: hot-metal robot
Casting Area + SOP
Guards the casting area
The casting area and SOP models watch the hot zone during tapping and pouring, raise an alarm when a person steps inside the exclusion line, and record the event.
So whatSafety checks that today depend on a supervisor being present run on every round, every shift, with the evidence attached.
Use cases · See, decide, act
The robot sees, decides and acts in one loop
Three walkthroughs with a Ripik model on board.
Industry example: robotic arm over a belt
Big Rock DetectionRobotic arm over a belt
SeeOversize lump on the belt ahead of the crusher
DecideSize exceeds the set threshold for this line
ActStops the belt, alerts the control room, logs the frame
Industry example: quadruped patrol
Foreign Object DetectionQuadruped patrol
SeeMetal strap riding the feed belt toward the mill
DecideNot material: a foreign object that will damage the liner
ActFlags the object, triggers the diverter, calls the operator
Industry example: inspection drone
Flare StackInspection drone
SeeFlame shape and colour at the tip of the stack
DecideSmoking flare: combustion outside the normal band
ActRaises the alarm, records the event for the compliance log
So whatThe action is configured by you: stop, alert, divert, log. The seeing and the deciding come ready from Ripik.
For partners
Ready skills for the robots already walking, rolling and flying through heavy industry
Ripik works with the companies that build, integrate and automate robots for iron & steel, metals, cement, mining, energy and pharma plants.
Inspection robot makers
Your robots already walk, roll or fly through the plants we serve. Ripik's models give them ready skills for material, equipment and safety monitoring, proven across 300+ deployments in 150+ plants, so your heavy-industry customers get answers from every round.
Partner or integration call
Physical-AI and humanoid companies
Heavy industry needs robots that understand belts, ladles and kilns. Ripik brings models trained on that footage, ready to run on your platform.
Partner call
Industrial OEMs and integrators
Add a proven AI vision layer to the equipment you already sell into steel, cement and mining plants, and open Ripik's customer base to a joint offer.
Joint offer or pilot
Automation vendors
Ripik turns a camera feed into an event your controller, SCADA or edge platform can act on, so a detection becomes an interlock or an alarm in your stack.
Technology partnership
Indian robotics companies
Ripik already works with India's largest steel and cement producers. Together we can take robots with ready AI into those plants.
Joint go-to-market
Plant leaders
Operations, maintenance, safety and digital teams can run Ripik models on robots they already own or plan to buy. A 4-week pilot shows whether a robot with Ripik models catches real problems on your site. Contact us to get started →
Summary
Plug Ripik models in and your robot acts
Three things to take away, and three steps to start.
3
levels of a robot
move, see, act. Level 3 is our expertise
42
ready-to-load models
iron & steel, metals, cement, mining, energy, pharma
Plug & play
no training needed
load the models and go
01
Your robot already moves. We make it see and act.
The hardware is a solved purchase. The judgement behind the camera is what turns a patrol into a response.
02
The models arrive trained and ready to load.
Each of the 42 use cases is already trained on real plant footage. Plug it in, with no labelling, data-science team or re-training per site.
03
Any carrier, any mix of use cases.
Quadruped, rail, drone, AGV or arm. Pick the models for the rounds your robot walks and add more as they grow.
How to start
1Pick the first use cases
2Choose or bring your robot
3Load the models and run the first round
So whatRipik's 42 ready models on your robot, in weeks, with no training needed.
Get started
Contact us to get started
Interested in running Ripik models on your robot, or in partnering with us? Tell us about your robot and the use cases you have in mind, and we will come back to you.
Start herePilot: 4 weeks·up to 3 use cases·on your robot, or on a partner platform we bring·fixed fee, credited if you continue
Thank you. Your enquiry has been received.
We will come back to you shortly to pick the first use cases and set up a call.
Ripik's 42 ready models on your robot, in weeks, with no training needed.
Big Rock Detection
Furnace Monitoring
Flare Stack
Coil Grading
OSHA Compliance
Foreign Object Detection
