Physical AI for the work robots still can’t do
We build the soft actuation, the real-time force control, and the AI that runs on them. Robin is what happens when all three are designed together: a two-armed mobile robot that works safely beside people, without a cage.
The robot arm is the cheap part.
The fencing, the fixtures, the guarding and the line rebuild around it are what cost you — and they’re what make the cell impossible to move once it’s built.
Even inside the fence, the economics break down. Every jam, drop or reset pulls a worker off their own job. A robot that can’t recover on its own doesn’t remove labor; it relocates it.
Roughly a tenth of the force
In our own comparative testing Robin generates roughly a tenth of the impact force of a conventional rigid arm under equivalent contact conditions. That’s the engineering basis for working without fencing — not a software speed limit.
Illustrative diagram of the ratio, not test data. Both bars rise with contact speed; Robin’s stays roughly a tenth as high, and below the limit a person can safely absorb. Certified figures follow formal testing.
Three layers, one company
Most robotics companies assemble off-the-shelf actuators, third-party perception, and a rented foundation model — and leave the customer to manage the seams between them. We build all three layers ourselves, because the thing that makes Robin work only exists where they meet.
Soft actuation
Compliant force-controlled actuators of our own design, mechanically soft by construction rather than by software. Four US patents granted. Compliance is mechanical, not simulated in software.
Real-time force control
Full-time closed-loop force control. Robin commands force directly rather than commanding a position and absorbing whatever force results. This is what the AI gets to build on.
Physical AI
A force-aware model trained in-house, plus the fleet software that maps, navigates, records demonstrations, trains policies and runs them. One system, one update path.
A fix ships as a single update, not an integrator project. A new capability arrives on the robot you already have. And because the same team owns the actuator and the model, the model can learn from a signal that most robots cannot even produce.
A specialist model, not a general-purpose one
General-purpose robot foundation models and our model solve different problems. Ours is built for one job: reliable, contact-rich industrial manipulation. That narrowness is the point — it’s how the model reaches production-grade reliability on a fraction of the data and compute.
Force is a training modality, not a safety cutout
Our model learns from force alongside vision, with force sensing at every joint feeding the policy directly. Insertion, hand-offs and gentle placing succeed because the model reacts to what the arms feel, not only to what the cameras see. It’s a modality most models ignore, because most robots can’t generate it.
Around a hundred demonstrations, not internet-scale datasets
A floor operator teaches a new task in VR while the system records episodes. Processing, training and deployment run from the same browser panel. New tasks arrive in days, without fleet-scale data collection and without an integrator.
Learned skills inside deterministic execution logic
The policy handles the contact-rich part; behavior trees handle sequencing, retries and recovery. Autonomy stays bounded and auditable rather than open-ended, which is what a production line actually needs — and what lets the robot recover from a fault without pulling an operator over.
Real-time on the robot, with no cloud in the control loop
The full stack runs on the machine. Deploy cloud-connected, on-premise, or fully air-gapped. No external dependency between sensing and acting, which is what makes it deployable inside enterprise IT and safety review.
| Dimension | Generalist foundation models | Our model |
|---|---|---|
| Purpose | One model for every robot and task | Purpose-built for industrial manipulation |
| Optimized for | Breadth of capability | Production-grade task reliability |
| Contact handling | Vision-first, position-centric | Force-native — feels and reacts to contact |
| Data required | Internet-scale datasets | Days of operator demonstrations |
| Compute | Data-center class | Real time, on the robot |
| IP position | Openly published | Proprietary, developed in-house |
The model, the training pipeline and the actuation it controls are developed under one roof. Capabilities reach customers as software updates on the robot they already have.
Robin, in numbers
Factory-default configuration. Two 7-DOF arms on an omni-directional base, with a motorized lifting column that matches human workstation heights. We publish these openly — you shouldn’t have to email someone to find out whether a robot fits your station.
Indoor, dry use. 10–32 °C, IP20. Default parallel gripper with 150 mm stroke and swappable contact surfaces; a five-finger hand is available for dexterous work. Design references ISO 10218-1 and -2:2025, ISO 3691-4, ISO 12100, ISO 13849-1, IEC 62061 and IEC 60825-1 Class 1. Product safety certification is underway for each target market; units deployed before certification operate under a service agreement with Roboligent-trained personnel.
Contact-rich handling work
The capabilities are primitive skills that compose into workflows, so one robot covers many processes on the same site.
Logistics and fulfillment
The beachhead application. Robin works at live pick stations alongside human workers.
- Tote and container handling between conveyors, racks and carts
- Item picking out of totes — mixed SKUs, varied geometry
- Heavy and irregular items, including non-rigid shapes
- Station-to-station transport with autonomous navigation
Automotive and manufacturing
Sequencing, kitting and line-side work where parts are heavy, varied, and sit in fixtures that demand contact awareness.
- Part sequencing from bin to rack, including pigeonhole insertion
- Mechanical assembly steps on engine-block-class components
- Line-side delivery and replenishment across shared floor space
- Fitting tasks where force feedback decides success, not vision
Electronics, display and semiconductor
Chemical handling — picking a bottle, opening a sealed cap, pouring powder to a target weight by the gram. Precision placement where excessive grip force damages the part.
Heavy and awkward payloads
Squeezing palms together to lift large boxes by friction, with no purpose-built gripper. Hugging items between both arms up to 20 kg. Pushing and pulling loads on and off an onboard lift cart.
You don’t buy a robot. You buy the work
Robin is provided under a subscription service agreement rather than a one-time sale. No capital purchase, no integration contract, no depreciating asset if the task changes.
Pilot
A robot on one station, on your floor, on your parts. Fixed scope and agreed success criteria before anyone signs anything longer.
Deploy
The station goes into production under a subscription term. Our technicians install and commission; your operators are trained and on record.
Expand
Add robots or move them to harder tasks as the mix changes. Every step is the same robot on a new job, not a new integration project.
What the subscription covers
- Robin hardware and the accessories for the defined deployment
- Preventive and corrective maintenance, parts and labor included
- Fleet management software, updates and security patches
- Firmware updates for control, safety and subsystems
- Remote support, with on-site visits when remote diagnosis isn’t enough
- Installation and commissioning of the initial deployment
Fault codes and safety events are forwarded to our service team automatically. For registered units a ticket opens when the fault is logged — nobody has to file anything.
Where we are
Founded 2016 in Austin, Texas, out of research at UT Austin. Hardware, control software and the AI stack are built in-house — there’s no third-party seam for a customer to manage.
Three on our compliant actuator and its internal assemblies, plus a granted design patent. Counterparts allowed in Europe, Japan, Korea and Canada; a gripper patent is allowed and awaiting issue.
Phase II contract from AFWERX, the U.S. Air Force innovation arm, with UT Austin as research partner — applying the same soft actuation and force control to physical rehabilitation.
An earlier National Science Foundation SBIR award backing the core actuation work.
Actuators, control software, the AI model and the fleet platform are all our own work, developed under one roof.
Designed, built and supported in Austin, Texas. Robin is assembled at our own facility in Round Rock, just outside Austin, by the same team that designs it. Service and engineering support come from that building, not from a distributor three time zones away.
Five open roles
Hardware, controls and AI sit in the same building in Round Rock. Everyone here touches something that ships to a customer.
Tech Lead, Robot Management System Round Rock, TX · On-site
Own the software that runs a fleet: mapping, navigation, task orchestration, demonstration recording and the browser panel operators work in. You will set the architecture and write the core of it, working next to the people who build the actuators and the control stack.
Senior, hands-on. C++ and Python, ROS 2, distributed systems.
Apply on LinkedInTech Lead, Robot Control Round Rock, TX · On-site
Real-time force control on our own hardware. You will own the control loop from actuator firmware to whole-body coordination, and the safety behavior that lets a robot work beside a person without fencing.
Senior, hands-on. Real-time C++, impedance and force control, robot dynamics.
Apply on LinkedInAI Engineer, Robot Data and Evaluation Round Rock, TX · On-site
Own the data loop behind our manipulation model: how demonstrations are captured and curated, how policies are trained, and how we decide whether a policy is good enough to run on a customer floor.
Python, PyTorch, imitation learning. Evaluation discipline matters more than model novelty here.
Apply on LinkedInField Integration Engineer Round Rock, TX · On-site, with travel
Take a robot to a customer station and make it work there. Scope the task, commission the cell, teach the policies, train the operators, and bring back what development got wrong. Roughly monthly domestic travel, twice yearly international.
Senior individual contributor, growing into field team lead. Comfortable being the only Roboligent person on site.
Apply on LinkedInManufacturing Quality Engineer Round Rock, TX · On-site
Build the quality function for our pilot production run: supplier quality as the anchor, incoming inspection, work instructions, and the yield data that tells us whether a design is ready to ship.
Hardware quality in low-volume, high-complexity assembly. First quality hire, so you are setting the process, not inheriting it.
Apply on LinkedInEach role is also posted on LinkedIn. Don’t see yours? Tell us what you’d build here — info@roboligent.com
Bring us one station you’ve been told can’t be automated
If it’s high-mix, hard to fixture, or too close to people to fence, that’s the one we want to see. Send a photo and a cycle time and we’ll tell you honestly whether Robin is the answer.