Franka Research 3 Review: The Lab Default, and Its Corporate Asterisk

If you have read a robot-learning paper in the last few years, you have seen this arm. The Franka Research 3, and the Panda before it, became the default manipulator in academic robotics for one concrete reason: joint torque sensing on all seven axes, exposed through a real-time control interface at 1 kHz. That combination lets a researcher write force and impedance controllers directly, which most industrial arms simply do not permit.

It is an excellent research platform. It is also sold by a company that went through insolvency, changed hands, and left an unresolved dispute over its patent portfolio, which is worth understanding before a lab commits to it as a decade-long standard.

Companion reading: collaborative robots (cobots), force and torque sensing, imitation learning for robotics, reinforcement learning for robotics, and how to choose a cobot.

Table of contents

1. Key takeaways
2. Why labs standardise on it
3. The specs, and the 3 kg ceiling
4. The corporate asterisk
5. How it compares
6. Who should buy it
7. Verdict
8. Frequently asked questions
9. Changelog

Why labs standardise on it

Most collaborative arms give you a position interface and a friendly teach pendant. That is the right design for a factory and the wrong one for a manipulation researcher, because the interesting behaviours (compliant insertion, contact-rich assembly, learned policies that modulate force) live below the level those interfaces expose.

The FR3 exposes that level. Torque sensors integrated at every joint estimate external contact forces, a fast control loop reacts promptly to collisions, and the Franka Control Interface runs at 1 kHz for torque commands and sensor readback. You can implement an impedance controller and mean it.

Around that sits the ecosystem that matters for a lab: ROS and ROS 2 integration, MATLAB and Simulink support, and enough low-level access that published work is reproducible on someone else's unit. The network effect is real. Standardising on the arm your field publishes against saves months.

The seventh axis is the other quiet advantage. Redundancy means the arm has a null space: it can hold the end-effector still and reconfigure its elbow, which matters for reaching into constrained spaces and for whole-arm collision avoidance. For where that sits in the wider picture, see collaborative robots.

The specs, and the 3 kg ceiling

The 3 kg payload is the number that decides whether this arm belongs in your project, and the same warning from any cobot applies with more force here: that budget covers the gripper, its mounting, cabling and the part. With a Franka Hand or a third-party gripper mounted, the workpiece allowance is small. This is an arm for manipulating objects you can pick up with two fingers, not for moving stock.

Read that as a scope statement rather than a flaw. The FR3 is designed for research and teaching and prototyping, and 3 kg is enough for essentially every published manipulation benchmark. It is not enough for production material handling, and no amount of clever control changes that.

The ±0.1 mm repeatability is worth flagging because it surprises people who assume research-grade means imprecise. It does not.

The corporate asterisk

This is the part a lab buying a ten-year platform should know, and it appears in no product page.

Franka Emika filed for insolvency in August 2023, having failed to secure follow-up financing, citing irreconcilable shareholder differences. In November 2023, roughly two months later, Agile Robots AG acquired the company in a deal reported at more than €30 million. Agile Robots retained the roughly 100-person team and stated plans to expand the portfolio and continue R&D. The business trades as Franka Robotics today.

Two complications sit around that:

1. Roughly 700 patents were sold a few weeks before the insolvency filing. The insolvency administrator worked to reverse that sale, and shareholders obtained preliminary injunctions against the purchase agreement. The final position on those steps has not been clearly reported.
2. The Munich I public prosecutor's office opened an investigation into the startup for subsidy fraud.

What this means practically, stated fairly: the product is alive, staffed, sold and supported, and an acquisition by a well-funded competitor is a considerably better outcome than liquidation. The FR3 is not orphaned hardware.

It does change the diligence questions. If you are standardising a lab on this arm for a decade, ask about the spare-parts commitment, the support horizon, and the roadmap owner, and get the answers in the quote rather than from a press release. Hardware that outlives its vendor is a familiar problem in robotics, and the cost lands years later when a joint module needs replacing.

How it compares

Against Universal Robots the split is clean: UR is the production arm your technicians can re-task, the FR3 is the research arm your students can write controllers for. Buying either for the other's job goes badly.

Who should buy it

Buy the FR3 if you run a manipulation research group, you need real-time torque control, and you want to reproduce and be reproduced by the published literature. There is no better-supported default for that work.

Buy it for teaching if your students should touch force control rather than only waypoints. The redundancy and the compliance behaviour make concepts tangible in a way a position-controlled arm cannot.

Do not buy it for production. 3 kg after tooling, at research-platform pricing, with a support structure in transition. That is the wrong tool and the wrong risk profile for a line.

Verdict

The FR3 earned its position honestly. Seven axes, torque sensing at every joint, a genuine 1 kHz control interface, ±0.1 mm repeatability, and an ecosystem where the papers you want to reproduce were written on the arm you own. For manipulation research it remains the default, and the default is worth something in itself.

Go in with the payload understood as 3 kg after tooling, and with the corporate history on the table. Insolvency, acquisition, a disputed patent sale and a fraud investigation is an unusual amount of turbulence for a platform purchase, and the mitigating fact is real: it was bought by a competitor with money who kept the team. Ask about parts and support horizon in writing, then buy it for the research it is genuinely the best tool for.

Frequently asked questions

What makes the Franka Research 3 different from a normal cobot? Torque sensors at every one of its seven joints, exposed through the Franka Control Interface at 1 kHz. That allows real torque and impedance control, which most industrial cobots do not expose. It is why the arm is the default in manipulation research.

How much can the FR3 lift? 3 kg over an 855 mm reach, and that budget must cover the gripper, its mounting and any cabling before the workpiece. It is a research and prototyping envelope rather than a production one.

Is the FR3 accurate? Yes. Pose repeatability is quoted at ±0.1 mm, which is industrial-grade despite the research positioning.

Why does the seventh axis matter? Redundancy. A 7-DoF arm can hold the tool pose fixed while reconfiguring its elbow, which helps reach into constrained spaces and avoid obstacles with the whole arm. A 6-axis arm has no such freedom.

What happened to Franka Emika? It filed for insolvency in August 2023 after failing to secure follow-up financing, citing irreconcilable shareholder differences. Agile Robots AG acquired it in November 2023 in a deal reported above €30 million, retained the roughly 100-person team, and the business now trades as Franka Robotics.

Is it risky to buy one after the insolvency? Less than it sounds, with caveats. The product is staffed, sold and supported under new ownership, which beats liquidation. But roughly 700 patents were sold weeks before the filing and that sale was disputed by the administrator and by shareholders, and a subsidy-fraud investigation was opened. For a decade-long platform decision, get the spare-parts and support-horizon commitments in the quote.

FR3 or Universal Robots? Different jobs. UR is a production arm optimised so your own technicians can re-task it. The FR3 is a research arm optimised for low-level control. Buying either for the other's role goes badly.