Fourier Intelligence GR-1 Review: Specs, AI & Price

Fourier Intelligence GR-1: Quick Buyer Verdict

The Fourier GR-1 should be evaluated as a human-scale humanoid development platform, not as a finished general-purpose employee. Its value comes from combining a commercially developed bipedal body, high-torque joints, seven-degree-of-freedom arms, interchangeable end-effector configurations and a software stack that supports secondary development, simulation and research integration.

For a research team, the GR-1 remains technically relevant. Fourier’s official documentation publishes a walking speed of 5 km/h, a maximum joint peak torque of 230 N·m and a single-hand load of approximately 3 kg. Its onboard computer is documented with an Intel Core i7-13700H, 16 GB RAM and 512 GB SSD in legacy GR-1 documentation, with Ubuntu 20.04 and ROS 2.

The complication is purchasing in 2026. GR-1 is no longer Fourier’s newest humanoid generation, and the current developer ecosystem uses naming that does not perfectly match the original product brochure. That makes configuration verification more important than the model name alone.

Pros

• Human-scale 1.65 m form factor for realistic bipedal and manipulation research.
• Up to 230 N·m published peak joint torque.
• Seven-degree-of-freedom arms and three-degree-of-freedom waist.
• Dexterous-hand and gripper/jaw variants.
• Published single-hand load of approximately 3 kg.
• Secondary development is supported in the product literature reviewed.
• Official URDF/model resources and support for NVIDIA Isaac Gym and Isaac Sim.
• Fourier Aurora SDK includes Python tooling and MuJoCo integration.
• Independent research has used GR-1 for locomotion, teleoperation and embodied-AI experiments.
• GR-1 was used as a physical platform in NVIDIA GR00T N1 research.

Cons

• No simple current manufacturer MSRP for a normal commercial GR-1 purchase was found in the materials reviewed.
• GR-1 is no longer Fourier’s newest humanoid platform.
• Official documents from different revisions publish different weight, arm-span and battery figures.
• The current Aurora SDK names GR-1P rather than generic GR-1, so compatibility must be checked for older units.
• The approximately 3 kg single-hand load is modest for many industrial handling tasks.
• Legacy official documentation lists only about 45 minutes of walking endurance.
• Advanced autonomy requires substantial software, training data, compute and robotics expertise.
• Fourier’s operating documentation shows that physical setup and standing procedures require controlled handling.
• No public specification reviewed establishes unrestricted collaborative operation around people.
• Buying an older-generation humanoid without a written support and spare-parts plan creates lifecycle risk.

Our recommendation: shortlist the GR-1 when you specifically need a human-scale Fourier platform for research, embodied AI, teleoperation or manipulation and can verify the exact configuration. Do not buy on the model name alone. Require the quote to identify the body revision, end effector, actuator and firmware versions, compute, sensors, battery revision, SDK compatibility, included software, spare parts, support term and acceptance tests. Review the Fourier Intelligence GR-1 listing before making an enquiry.

How Much Does the Fourier Intelligence GR-1 Cost in 2026?

Fourier does not publish a simple current general-market GR-1 MSRP in the manufacturer materials reviewed for this article. That makes a configuration-level quote more important than a headline number.

There is public 2026 pricing for a Fourier GR-1 Competition Version through the RoboCup Humanoid Robot Program. The programme lists a regular price of US$200,000 and a special offer price of US$50,000 for one GR-1 for eligible teams. Those figures belong to a competition programme with specific participation conditions and should not be treated as the standard commercial street price of every GR-1 configuration.

The robot price is not the programme price

A useful GR-1 research deployment can require significantly more than the humanoid body itself.

A better way to request a GR-1 quote

Ask for two figures:

1. Minimum experiment configuration: the exact robot, end effectors, battery, controller, development access and compute required to run one clearly defined experiment.
2. Two-year programme cost: hardware, freight, duties, spare parts, batteries, support, repairs, safety equipment, compute, teleoperation equipment and engineering labour.

Do not compare a programme-specific GR-1 price with another humanoid’s bare chassis price. Compare equivalent hardware, manipulation capability, developer access, support and warranty.

What Is the Fourier Intelligence GR-1?

The Fourier Intelligence GR-1 is a general-purpose humanoid robot platform developed for humanoid locomotion, manipulation, research, education and embodied-AI development. Fourier introduced the GR-1 generation before its newer GR-2 and GR-3 systems, but the platform remains documented and continues to appear in official model repositories and research projects.

Depending on the source revision and configuration, the GR-1 is approximately 1.65 m tall and roughly 52–60 kg. Fourier’s current legacy developer documentation lists a 5 km/h walking speed, approximately 3 kg single-hand load and 230 N·m maximum joint peak torque.

The main body architecture includes:

• Three head degrees of freedom in legacy GR-1 documentation.
• Three waist degrees of freedom.
• Seven degrees of freedom per arm.
• Six degrees of freedom per leg.
• Dexterous-hand or jaw/gripper end effectors depending on variant.

What the GR-1 is

• A human-scale physical platform for bipedal locomotion research.
• A secondary-development platform for robotics teams.
• A platform that can be used with simulation and reinforcement-learning workflows.
• A body for teleoperation and humanoid-data collection.
• A research platform for bimanual manipulation and embodied AI.
• A useful bridge between simulation-only humanoid work and real full-body hardware.

What the GR-1 is not

• It is not Fourier’s newest humanoid generation in 2026.
• It is not a turnkey autonomous worker that arrives knowing arbitrary tasks.
• It is not a heavy-lifting humanoid based on the current 3 kg single-hand specification.
• It is not automatically safe for unrestricted operation beside untrained people.
• It is not an all-weather outdoor platform based on the public documentation reviewed.
• It is not one perfectly uniform hardware specification across every document and unit.
• It is not proof that a research demonstration will transfer directly to a production workflow.

If you are still comparing the category, browse current humanoid robots, the best humanoid robots in 2026 and the Fourier Intelligence brand page.

GR-1 vs GR-1L vs GR-1 Pro vs GR-1L Pro

One of the easiest ways to misunderstand the GR-1 is to treat “GR-1” as one fixed hardware package. Fourier’s official robot-model repository maps four commercial names to two main body codes and two end-effector families.

The official model repository also warns that there are different dexterous-hand types and tells developers to select the URDF that matches the physical hand.

Why this matters more than the marketing name

A controller or policy depends on the robot’s actual kinematic chain. A GR1T1 with a Fourier hand is not necessarily interchangeable with a GR1T2 using another end effector. The model file, joint limits, hand interface, calibration and software support all need to match the physical robot.

A serious quotation should therefore identify:

• Commercial product name.
• Body code or equivalent hardware revision.
• End-effector make and model.
• Number of actuators and controlled joints.
• Sensor configuration.
• Onboard computer.
• Battery model and capacity.
• Actuator communication and driver firmware.
• Aurora SDK compatibility.
• ROS and simulation resources.
• Warranty, spare parts and support period.

What is GR-1P?

The current Fourier Aurora SDK v1.3.0 lists GR-1P, GR-2, GR-3 and Fourier-N1 as supported robots. By contrast, the older GR-1 product documentation and model repository use GR-1, GR-1L, GR-1 Pro, GR-1L Pro, GR1T1 and GR1T2 naming.

That nomenclature mismatch should be treated as a purchasing question, not guessed away. Ask Fourier or the supplier to confirm whether the exact GR-1 unit being quoted is supported by the current Aurora release, which configuration files apply and whether any actuator or firmware upgrade is required.

Fourier Intelligence GR-1 Specifications

The table below prioritises Fourier’s official GR-1 developer documentation and later GR-1 family brochure. Where those sources disagree, the difference is identified in the next section rather than selecting one value without explanation.

Does the GR-1 have 44 or 54 degrees of freedom?

Both figures appear in Fourier materials because documents use different ways of describing the platform and its end effectors. The official legacy developer page describes up to 44 degrees of freedom and also publishes 44/34 total actuators depending on configuration. A later GR-1 family brochure markets up to 54 degrees of freedom while separately listing actuator totals.

The safe buyer interpretation is simple: do not use the headline DoF number to infer the exact controlled joint chain. Require a joint schedule for the quoted robot and match it to the correct URDF.

Can the Fourier GR-1 carry 50 kg?

Do not use an early headline carrying claim as the payload specification for a 2026 purchase. Fourier’s current GR-1 technical documentation publishes a single-hand load of approximately 3 kg.

A whole-body demonstration, theoretical carrying claim, peak load or early promotional target is not equivalent to a repeatable per-arm payload rating. For manipulation projects, test the exact object, posture, reach, speed and duration.

Why Do Some Fourier GR-1 Specifications Conflict Online?

This is one of the most important findings in this review.

Fourier’s official GR-1 documentation has evolved across product generations, body revisions and documentation systems. As a result, legitimate Fourier sources publish values that do not always match exactly.

These differences do not automatically mean one source is “wrong.” They can reflect revised hardware, different variant definitions, changed end effectors or documentation updates.

The correct purchasing rule

Treat the delivered serial-number configuration as the product.

Before purchase, request:

1. A signed configuration schedule.
2. The exact mechanical drawing for the quoted revision.
3. The battery part number and capacity.
4. The exact URDF/MJCF or model package that matches the robot.
5. The supported Aurora version.
6. Required actuator firmware versions.
7. The end-effector model and interface.
8. Acceptance-test limits for speed, payload and runtime.

This is more reliable than combining numbers from different GR-1 webpages into one “perfect” specification sheet that may not describe any single physical unit.

Locomotion, Balance and Mobility

Locomotion is one of the strongest reasons to consider the GR-1. Fourier designed it as a full-size biped rather than a wheeled service robot with humanoid arms.

The official documentation publishes a walking speed of 5 km/h, up to 230 N·m peak joint torque and six degrees of freedom per leg. Fourier describes capabilities including fast walking, slope traversal and disturbance resistance, while independent researchers have used GR-1 bodies for learned locomotion and whole-body teleoperation.

Why human scale matters

At approximately 1.65 m tall, the GR-1 can interact with environments built for standing adults in ways that compact humanoids cannot always reproduce. That can matter for:

• Human-height shelves and benches.
• Door and appliance interaction.
• Bimanual tasks at adult work surfaces.
• Teleoperation using human motion.
• Whole-body manipulation.
• Research into human-compatible workspace geometry.

The trade-off is safety and handling. A 50-plus-kilogram humanoid is harder to catch, lift, support and protect than a compact research platform.

Can the GR-1 walk autonomously?

The robot can support walking control and developer modes, but this is different from unrestricted autonomous navigation.

A complete autonomous mobile system still needs to solve or integrate:

• Environment perception.
• Localisation and mapping.
• Path and footstep planning.
• Obstacle avoidance.
• Whole-body collision checking.
• Recovery from tracking or perception failure.
• Safe behaviour around people.
• Task-level planning.

Do not translate “can walk” into “can autonomously work anywhere.”

Can the GR-1 climb stairs?

Fourier has promoted the GR-1 as a humanoid capable of handling human environments, but the public technical documentation reviewed for this article does not provide one universal stair geometry that a buyer can treat as a guaranteed acceptance specification.

If stairs matter, define:

• Riser height.
• Tread depth.
• Stair width.
• Landing geometry.
• Surface friction.
• Presence of handrails.
• Lighting.
• Required carrying load.
• Recovery space.

Then witness the exact quoted robot completing that staircase.

The protection stand is not a minor detail

Fourier’s GR-1 operating guide states that the robot must be fixed on a protection stand when returning to its initial state. The operator then lowers the stand until the feet contact the ground and checks stability.

The same guide warns against leaving the robot standing on the ground for a long period because the ankle actuators may overheat or burn out.

That tells buyers something important about practical operation: the GR-1 should be treated as research equipment with defined setup, support and shutdown procedures—not as an appliance that can simply remain standing indefinitely.

Fourier GR-1 Arms, Hands and Manipulation

The GR-1’s upper body is one of its major advantages for embodied-AI research. Fourier documents seven degrees of freedom per arm, and its product family includes both dexterous-hand and jaw/gripper variants.

The published single-hand load is approximately 3 kg. That is enough for many research objects, tools, containers and household-scale manipulation experiments, but it is not a heavy industrial payload.

Dexterous-hand configurations

Fourier’s legacy GR-1 documentation describes its dexterous hand as a biomimetic end effector with six motion degrees of freedom. The official model repository also distinguishes between Fourier and Inspire hand configurations and instructs developers to select a matching robot model.

That matters because hand choice changes:

• Kinematics.
• Grasp strategy.
• Control interface.
• Available sensors.
• Object size and geometry limits.
• Teleoperation mapping.
• Dataset compatibility.

GR-1L jaw/gripper configurations

The GR-1L and GR-1L Pro use a Fourier jaw rather than the dexterous-hand configuration shown for GR-1 and GR-1 Pro in the official model repository.

A simpler gripper can be the better engineering choice when the task needs repeatable opening and closing rather than anthropomorphic finger motion. More fingers do not automatically create a more reliable application.

Can the GR-1 perform useful manipulation?

Yes, as a research platform. The GR-1 has been used in modern humanoid-AI research involving bimanual manipulation and teleoperation.

But useful manipulation is a system capability, not a body specification. A successful task may depend on:

• The exact hands or jaws.
• Camera placement.
• Calibration.
• Perception models.
• Demonstration data.
• Policy training.
• Offboard GPU compute.
• Object presentation.
• Safety constraints.

Buyer rule: define the object before choosing the robot. Record its mass, dimensions, surface, fragility, pickup pose, required orientation, reach, placement tolerance and cycle time. Then test those conditions on the quoted GR-1 configuration.

AI and Autonomous Capabilities

The GR-1 is best understood as AI-capable humanoid hardware. It provides a physical embodiment on which locomotion, manipulation, teleoperation and learned policies can be developed.

It should not be interpreted as a robot that already contains general-purpose intelligence capable of reliably performing any spoken task.

Why NVIDIA GR00T matters

The GR-1 gained important research relevance when NVIDIA used it as one of the physical humanoid embodiments in its Isaac GR00T N1 work. GR00T N1 is an open foundation model for humanoid robots, and the research includes language-conditioned bimanual manipulation on a Fourier GR-1.

This is meaningful evidence that the GR-1 can serve as a real platform for modern vision-language-action and humanoid foundation-model research.

It is not evidence that every GR-1 sold automatically includes GR00T, NVIDIA compute, trained skills or production-ready autonomy.

What the robot provides

Depending on configuration and software revision, the GR-1 platform can provide:

• Whole-body actuation.
• Joint-state feedback.
• Depth sensing.
• IMU data.
• Networked actuator control.
• Developer interfaces.
• Simulation models.
• Teleoperation integration.

What the buyer still has to build

For a new autonomous application, the team may still need:

• Task demonstrations.
• Training datasets.
• Perception and object tracking.
• Language or multimodal planning.
• Locomotion policies.
• Manipulation policies.
• Whole-body coordination.
• Failure detection.
• Safe fallback behaviour.
• Evaluation and monitoring.
• Retraining and version control.

Can the GR-1 use a large language model?

Yes. A robotics team can integrate language or multimodal models into the application stack. A language model can interpret instructions, reason over task context or select skills.

It does not replace real-time balance, collision avoidance, grasp control or safety. A physical robot still needs lower-level systems that verify and execute actions within controlled limits.

Marketing demonstration vs deployed capability

For every impressive GR-1 video, ask five separate questions:

1. Was the task teleoperated, scripted, learned or autonomous?
2. Which GR-1 hardware revision was used?
3. Which sensors and computers were added?
4. How many training demonstrations or trials were required?
5. What was the measured success rate under the buyer’s actual conditions?

A video proves that a system completed a demonstration. It does not by itself establish repeatability, autonomy, safety or commercial economics.

SDK, ROS, Simulation and Developer Ecosystem

The software ecosystem is one of the strongest reasons to evaluate the GR-1 as a research platform rather than only as a piece of hardware.

Operating system and ROS

Fourier’s legacy GR-1 documentation lists:

• Ubuntu 20.04.
• ROS 2.
• Intel Core i7-13700H.
• 16 GB RAM.
• 512 GB SSD.
• Wi-Fi and Bluetooth connectivity.

These values describe documented GR-1 hardware, but software images can change. Ask for the exact OS, kernel, ROS distribution, driver versions and upgrade policy for the unit being supplied.

Fourier Aurora SDK

Fourier Aurora stands for Advanced Unified Robot Operation and Resource Architecture. The current public SDK repository contains:

• Configuration files.
• A Python SDK.
• Example scripts.
• API documentation.
• MuJoCo simulation integration.
• A Docker workflow.

The current v1.3.0 repository lists GR-1P, GR-2, GR-3 and Fourier-N1 as supported robots.

This is why an older GR-1 should not be purchased with the vague promise that “it has an SDK.” Require the supplier to state which Aurora release supports the exact robot and whether firmware upgrades are necessary.

Python client and DDS communication

Fourier’s GR-1 Aurora documentation provides a Python client and describes DDS-based communication for commands and robot state.

That is useful for teams working in Python-based robotics and machine-learning stacks, but buyers should still determine:

• Which command modes are available.
• Which joint groups can be controlled.
• State-data frequencies.
• Network configuration.
• Real-time limitations.
• Actuator-level versus higher-level access.
• Safety limits imposed by the controller.

Robot models, Isaac Gym and Isaac Sim

Fourier’s official GRx model repository provides URDF resources and support material for simulation workflows. It explicitly mentions NVIDIA Isaac Gym and Isaac Sim and separates GR1T1 and GR1T2 model variants.

This is valuable for:

• Reinforcement-learning policy development.
• Kinematic testing.
• Collision modelling.
• Controller prototyping.
• Simulation-to-real transfer.

The main risk is model mismatch. A policy trained against one hand, mass distribution or joint definition can behave poorly on a different physical revision.

MuJoCo

The current Aurora SDK includes a MuJoCo simulation directory, and Fourier’s model repository points developers toward MJCF conversion resources.

Simulation reduces physical risk and iteration cost, but it does not eliminate:

• Contact-model mismatch.
• Actuator latency.
• Backlash and compliance.
• Sensor noise.
• Battery-state effects.
• Real floor friction.
• Network timing.

Every sim-to-real deployment should begin with conservative limits.

Teleoperation

Fourier also maintains public teleoperation software resources, and independent researchers have developed GR-1 teleoperation systems using exoskeleton hardware.

Teleoperation can serve two different purposes:

1. Direct operation: a human controls the robot to perform a task.
2. Data collection: human demonstrations are recorded to train imitation-learning or vision-language-action policies.

The second use case is increasingly important for humanoid AI because high-quality embodiment-specific demonstrations are expensive to collect.

Developer verdict

The GR-1 is appropriate for teams comfortable with Linux, ROS, Python or C++, networking, robot models, simulation, machine learning and physical safety.

A buyer without those capabilities should budget for a systems integrator or choose a more vertically integrated robot/application package.

Battery Life, Charging and Duty Cycle

Battery documentation is one area where GR-1 revision differences matter.

Fourier’s legacy developer documentation publishes:

• 483 Wh battery capacity.
• Approximately 60 minutes of endurance.
• Approximately 45 minutes of walking endurance.
• 315 minutes of charging time.
• 46 V, 2 A maximum adapter output.

A later Fourier GR-1 family brochure lists a 460 Wh battery. The correct conclusion is not to average those numbers. Confirm the battery installed in the robot being quoted.

What changes real runtime?

• Walking speed.
• Continuous standing.
• Dynamic motions.
• Arm and hand activity.
• Payload.
• Additional compute.
• Camera and perception workloads.
• Network activity.
• Battery age.
• Temperature.
• Required safety reserve.

Why 45 minutes of walking matters

A walking-endurance figure of approximately 45 minutes puts the GR-1 much closer to a research-session platform than an all-shift commercial worker.

That may be perfectly acceptable for:

• Laboratory experiments.
• Teleoperation studies.
• Short data-collection sessions.
• Controller testing.
• Demonstrations.

It is much less attractive for a workflow that expects hours of uninterrupted autonomous labour.

Plan around usable experiment time

Even a nominal 60-minute runtime does not equal 60 minutes of productive trials. Booting, calibration, debugging, standing, failed episodes and resets consume energy.

A good research plan should define:

• Minimum battery threshold for starting a trial.
• Low-battery behaviour.
• Charging rotation.
• Whether spare batteries are available.
• Battery storage and transport rules.
• Whether battery replacement requires shutdown.
• Battery health logging.

Do not assume hot-swapping unless the supplier confirms it for the exact GR-1 revision.

Fourier GR-1 Safety and Operating Limitations

The GR-1 is a powerful, human-size machine. Its scale and joint torque create credible impact, fall, pinch and crush hazards.

The operating guide itself demonstrates why controlled procedures matter. Fourier instructs users to initialize the robot on a protection stand, lower it carefully, check stability and support it during the standing process.

Minimum laboratory controls

• A defined fall and exclusion zone.
• Trained operators only during dynamic testing.
• A protection/support stand compatible with the robot.
• Clear emergency-stop ownership.
• Impact-tolerant or protected flooring where appropriate.
• Protected cameras, workstations and fragile equipment.
• A pre-start inspection.
• Known joint and speed limits.
• A procedure for lifting or recovering the robot after a fault.
• A battery and damaged-component process.

The ankle-overheating warning

Fourier’s operating guide warns not to leave the robot standing on the ground for a long time because the ankle actuators may overheat or burn out.

For a buyer, this has three consequences:

1. Continuous standing should not be assumed to be a zero-cost idle state.
2. Duty-cycle planning should include supported or powered-down periods where appropriate.
3. A commercial workflow should not be designed until thermal behaviour is tested under the intended standing and walking pattern.

Emergency-stop behaviour must be tested

The robot has an emergency-stop interface in Fourier’s electrical documentation. However, a buyer should not infer the full physical stopping behaviour from the existence of an E-stop button.

For a biped, loss of active control can itself create a fall hazard. Acceptance testing should establish:

• What the controller does when E-stop is activated.
• Whether the body remains supported or falls.
• Which joints remain powered.
• How the robot is restarted.
• Whether a payload can be dropped.
• How operators stay outside the fall envelope.

Ingress protection and outdoor operation

The public GR-1 specification reviewed for this article does not establish an IP rating that would justify assuming rain, washdown or dust protection.

Until the quoted configuration has written environmental limits, avoid assuming suitability for:

• Rain.
• Standing water.
• Washdown areas.
• Heavy dust.
• Extreme temperatures.
• Explosive atmospheres.
• Uncontrolled outdoor terrain.

Can the GR-1 work safely beside people?

The documents reviewed do not establish unrestricted collaborative operation around untrained people.

A workplace deployment needs a task-specific risk assessment covering:

• Speed and separation.
• Foreseeable falls.
• Pinch points.
• Payloads and tools.
• Stopping behaviour.
• Software failures.
• Network failures.
• Unauthorised access.
• Battery hazards.

Cybersecurity

The GR-1 is a networked programmable machine. Its documented architecture includes Ethernet networking, Wi-Fi, Bluetooth, an onboard computer and developer control interfaces.

Treat it like an enterprise robot, not an isolated mechanical device. Review:

• Network segmentation.
• Credentials and privileged access.
• Remote terminal access.
• Software and firmware update provenance.
• Developer-machine access.
• Command permissions.
• Logging.
• Incident response.

What Real-World Fourier GR-1 Research Shows

The strongest evidence for the GR-1 is not a marketing video. It is the fact that independent research teams and major AI platforms have used the robot as a physical embodiment.

What this evidence proves

• The GR-1 is a credible physical platform for advanced humanoid research.
• Its hardware can support learned locomotion and manipulation policies.
• It can be integrated into teleoperation and data-collection systems.
• It is relevant to current embodied-AI and humanoid foundation-model research.
• Its model ecosystem is useful enough for external researchers to deploy policies on real hardware.

What it does not prove

• That a newly purchased GR-1 includes those research capabilities.
• That the same policy will work on every GR-1 revision.
• That the robot can complete arbitrary tasks without training.
• That research success rates are sufficient for production.
• That the robot can work unattended for a full commercial shift.
• That a demonstration system used only the hardware included in a standard quote.

The pattern is the same across modern humanoid research: the robot body is only one layer. Useful capability comes from the complete stack of hardware revision, end effectors, sensors, compute, data, control policy and validation.

Best Uses for the Fourier Intelligence GR-1

1. Embodied-AI and humanoid foundation-model research

Best overall use case. The GR-1 is a human-scale body that has already been used in modern foundation-model research. Teams working on vision-language-action models, imitation learning or general manipulation can use the platform as a physical embodiment rather than developing a full humanoid from scratch.

2. Bipedal locomotion and whole-body control

The 1.65 m body, six-degree-of-freedom legs, three-degree-of-freedom waist and high peak joint torque make the GR-1 relevant for walking, balance, disturbance rejection and whole-body coordination research.

3. Simulation-to-real reinforcement learning

Official robot models and Isaac-compatible resources make the GR-1 suitable for teams training policies in simulation before controlled deployment on hardware.

The configuration must match the simulation model closely enough for the transfer to be meaningful.

4. Teleoperation and robot-data collection

The GR-1 can be used with whole-body teleoperation systems to collect human demonstrations. This is particularly valuable for imitation learning and vision-language-action training, where robot-specific data can be a major bottleneck.

5. Bimanual manipulation research

Seven-degree-of-freedom arms and dexterous-hand variants create a useful platform for two-handed object handling, reaching, handovers and coordinated loco-manipulation.

The approximately 3 kg single-hand load limits heavier applications.

6. University robotics research

The GR-1 can support teaching and research in:

• Kinematics.
• Dynamics.
• Control.
• ROS.
• Reinforcement learning.
• Computer vision.
• Human-robot interaction.
• Teleoperation.
• Robot safety.

Physical access should still be restricted to trained users, with simulation used before hardware deployment.

7. Human-robot interaction experiments

Its human-scale form can be useful when the research question depends on body height, reach, proxemics or interaction with human-designed furniture.

That does not mean it is automatically safe for open public interaction.

8. Controlled technology demonstrations

The GR-1 has strong visual impact for laboratories, exhibitions and technology showcases.

Use bounded routines, controlled flooring, trained operators and audience separation.

9. Pre-industrial workflow prototyping

A GR-1 can help answer whether a humanoid form provides an advantage for a specific human-scale workflow.

Use it to test the hypothesis. Do not assume a successful prototype is already a commercially reliable automation system.

When the Fourier GR-1 Is Not the Right Robot

The GR-1 should not be selected simply because a humanoid looks flexible.

• Heavy material handling: approximately 3 kg per hand is too low for many industrial lifting tasks.
• All-day autonomous work: legacy official endurance figures are far below a normal work shift.
• High-throughput fixed automation: an industrial arm or cobot will often be faster, easier to guard and more repeatable.
• Warehouse transport: an AMR can usually move much heavier loads with less energy and balance risk.
• Outdoor inspection: a weather-rated quadruped or tracked robot may be a better fit.
• Unrestricted public interaction: size, hard surfaces and fall risk make public deployment difficult without engineered controls.
• Plug-and-play general intelligence: the GR-1 does not arrive knowing arbitrary tasks.
• Teams without robotics engineers: the value of the platform depends on development and integration capability.
• Buyers who specifically want Fourier’s newest general humanoid hardware: compare the GR-2 first.
• Buyers focused on social-care interaction: compare Fourier’s GR-3, which is positioned around care and interaction rather than simply assuming GR-1 is the best Fourier option.

A humanoid earns its complexity when legs, arms and human-scale geometry solve a real constraint. If the task can be completed by a simpler robot, the simpler architecture will often be cheaper and easier to validate.

Fourier GR-1 Alternatives in 2026

The most important alternatives are not all direct substitutes. Some are newer Fourier platforms; others trade human scale for lower cost or a different developer ecosystem.

Which one should you choose?

• Choose GR-1 when you specifically value its existing research history, Fourier ecosystem and human-scale GR1T1/GR1T2 platform—and can verify support for the exact unit.
• Choose GR-2 when newer Fourier hardware is the priority.
• Choose GR-3 when interaction and care-oriented design are central to the project.
• Choose Unitree G1 EDU when compact size, lower published entry cost and a large current research ecosystem matter more than full human scale.
• Compare Unitree H2 when you want another current full-size humanoid rather than an older-generation research body.

Use the Anton Robots comparison tool to compare available humanoids by specifications and intended application.

Is the Fourier Intelligence GR-1 Worth It?

The Fourier GR-1 can still be worth considering in 2026 for a research team that needs a human-scale humanoid body, has a defined experiment and can obtain a fully supported configuration at a sensible programme cost.

Its value is strongest when the alternative is designing a complete bipedal body from scratch.

Where the value comes from

• Human-scale bipedal hardware.
• High joint torque.
• Seven-degree-of-freedom arms.
• Dexterous-hand and gripper configurations.
• Secondary-development support.
• Official robot models.
• Simulation integration.
• Existing independent research using the embodiment.
• Relevance to humanoid foundation-model development.

Where the risk comes from

• Older-generation hardware relative to GR-2 and GR-3.
• Configuration and naming complexity.
• Different official specifications across revisions.
• Current Aurora support naming that needs clarification for older GR-1 units.
• Limited published walking endurance.
• Modest manipulation payload.
• Potential spare-part and support lifecycle risk.
• Engineering effort required to create autonomous skills.

A practical value test

Before buying, complete this sentence:

We need the Fourier GR-1 specifically because our project requires ________, and the same objective cannot be achieved more effectively with simulation, a fixed arm, an AMR, a compact humanoid or a newer GR-series robot.

Good answers could include human-scale whole-body teleoperation, GR-1-specific policy research, reproduction of published GR-1 experiments, bimanual embodied-AI work or development against an existing GR1T1/GR1T2 software stack.

“We want a humanoid because humanoids are the future” is not enough to select a US$-tens-of-thousands research programme.

Fourier GR-1 Buying Checklist

1. Define the research or workflow objective. Write the exact task before choosing the robot.
2. Confirm the product status. Ask whether the unit is new production, existing stock, programme-specific hardware or refurbished/demo equipment.
3. Identify the body revision. Record GR1T1, GR1T2, GR-1P or the supplier’s equivalent hardware identifier.
4. Confirm the commercial variant. GR-1, GR-1L, GR-1 Pro or GR-1L Pro.
5. Identify the end effector. Fourier hand, Inspire hand, jaw/gripper or another tool.
6. Request the joint schedule. Do not rely on one headline DoF number.
7. Confirm payload conditions. Test the actual object at the actual reach and speed.
8. Confirm the battery revision. Obtain part number, Wh capacity, runtime, charging time and replacement availability.
9. Confirm compute. Record CPU, RAM, SSD, optional GPU/accelerator and offboard compute requirements.
10. Confirm perception hardware. Record camera model, IMU and any added cameras or microphones.
11. Verify Aurora compatibility. Record supported SDK version and required actuator/driver firmware.
12. Verify ROS and simulation. Match the physical robot to the correct URDF/MJCF and software versions.
13. Plan the safety area. Include a protection stand, fall zone, barriers and recovery method.
14. Test emergency behaviour. Establish what happens physically when E-stop or power loss occurs.
15. Request spare-parts availability. Include actuators, hands, batteries, cables and vulnerable external components.
16. Request support terms. Define response time, remote support, repair location and turnaround.
17. Define acceptance tests. Include boot, standing, walking, joint control, sensor streams, end effectors, runtime and SDK access.
18. Budget the programme. Include engineering labour, GPU compute, teleoperation, safety hardware, repairs and downtime.

Pro tip: require a live video acceptance test using the exact robot or an identically configured unit. A demonstration from a different GR-1 generation or hand configuration does not prove that the quoted system has the same software or capability.

How to Buy the Fourier Intelligence GR-1

Because GR-1 is no longer Fourier’s newest humanoid platform, the buying process should start with configuration and lifecycle questions rather than price alone.

A useful enquiry should state:

• Organisation and country.
• Research or commercial objective.
• Required body size and payload.
• Required end effector.
• Locomotion requirements.
• Required SDK and ROS access.
• Simulation environment.
• Teleoperation requirements.
• Onboard and offboard compute.
• Number of batteries.
• Expected delivery date.
• Required support duration.

Before paying, request:

• A formal configuration schedule.
• Hardware-revision identifier.
• Serial-number or production-batch information where available.
• Photographs and video of the exact configuration.
• Battery specification.
• Software and firmware versions.
• Aurora SDK compatibility statement.
• Correct robot-model files.
• Warranty terms.
• Repair and support process.
• Spare-parts availability.
• Delivery and import documentation.
• Acceptance-test procedure.

Review the Fourier Intelligence GR-1 product page, then contact Anton Robots to discuss requirements and available sourcing options. If the application is still unclear, use the Find My Robot tool before committing to one humanoid.

What Is New for the Fourier GR-1 in 2026?

GR-1 is now part of a multi-generation Fourier humanoid family

The biggest change is context. A GR-1 buyer is no longer comparing one Fourier humanoid with the outside market. Fourier now has newer GR-series platforms.

The GR-2 is the more obvious newer full-size comparison, while GR-3 is positioned as a care-oriented humanoid with a different interaction focus.

This makes “Why GR-1 instead of GR-2?” a mandatory purchasing question.

The current Aurora SDK uses GR-1P naming

Fourier’s public Aurora SDK v1.3.0 currently lists GR-1P, GR-2, GR-3 and Fourier-N1.

Older GR-1-specific documentation remains available, but this makes exact compatibility verification essential for any older GR1T1 or GR1T2 unit.

The GR-1 remains relevant to state-of-the-art research

NVIDIA’s GR00T N1 research and independent humanoid-control projects have kept the GR-1 relevant beyond its original launch cycle.

That is a stronger reason to buy it than nostalgia for an early humanoid launch. A research team may already have policies, datasets, robot models or papers built around this embodiment.

Documentation migration increases the need for buyer diligence

Fourier maintains a newer developer center while older GR-1 documentation remains in its legacy documentation system.

When buying in 2026, save the exact documentation and software versions that apply to the delivered robot rather than assuming future pages will continue to describe the same configuration.

Frequently Asked Questions

How tall is the Fourier GR-1?

The GR-1 is approximately 1.65 m tall according to Fourier documentation.

How much does the Fourier GR-1 weigh?

The answer depends on revision. Legacy developer documentation lists approximately 52 kg, while later GR-1 family material lists around 55 kg for base configurations and around 60 kg for Pro variants. Confirm the exact delivered revision.

How fast can the Fourier GR-1 walk?

Fourier publishes a walking speed of 5 km/h for the GR-1.

What is the GR-1 payload?

Fourier’s technical documentation publishes an approximately 3 kg single-hand load. Payload depends on posture, reach and end effector, so test the real object rather than treating 3 kg as a universal handling rating.

Can the GR-1 carry 50 kg?

Do not treat early promotional carrying claims as the current arm-payload specification. For purchasing, the relevant current GR-1 technical figure is approximately 3 kg per hand.

What is the maximum torque of the Fourier GR-1?

Fourier publishes a maximum joint peak torque of 230 N·m.

How many degrees of freedom does the GR-1 have?

Different Fourier materials use different headline counts, including up to 44 in legacy documentation and up to 54 in later product material. Buyers should request the exact joint and actuator schedule for the quoted configuration.

What is the difference between GR-1 and GR-1L?

Fourier’s official model repository maps GR-1 to a GR1T1 body with a dexterous hand and GR-1L to a GR1T1 body with a Fourier jaw.

What is the difference between GR-1 and GR-1 Pro?

The model repository maps GR-1 to the GR1T1 body and GR-1 Pro to the GR1T2 body, both with dexterous hands. Buyers should still request the exact mechanical and electrical differences for the quoted revision.

What is GR-1P?

GR-1P is the name currently listed as a supported robot in Fourier Aurora SDK v1.3.0. Because older GR-1 documents use GR1T1/GR1T2 and GR-1/GR-1 Pro naming, buyers should ask Fourier to map their exact hardware to the current SDK configuration.

Does the Fourier GR-1 support secondary development?

Yes, Fourier’s GR-1 family product material lists secondary development support. The exact software and firmware compatibility should be confirmed for the unit being purchased.

Does the GR-1 support ROS?

Fourier’s legacy GR-1 documentation lists Ubuntu 20.04 and ROS 2. Confirm the exact ROS distribution, drivers and maintained interfaces on the delivered software image.

Does the GR-1 have a Python SDK?

Fourier provides Python client tooling for its Aurora system. Current support should be verified against the exact GR-1 hardware and firmware.

Can the GR-1 run in MuJoCo?

Fourier’s current Aurora SDK includes MuJoCo simulation resources, and Fourier publishes robot model assets that can support simulation workflows.

Can the GR-1 use NVIDIA Isaac Sim?

Fourier’s official GRx model repository explicitly lists support resources for NVIDIA Isaac Gym and Isaac Sim.

Has the GR-1 been used with NVIDIA GR00T?

Yes. NVIDIA’s GR00T N1 research used a Fourier GR-1 for language-conditioned bimanual manipulation experiments.

Can the GR-1 be teleoperated?

Yes. Fourier publishes teleoperation resources, and independent research such as HOMIE has developed whole-body teleoperation systems for the GR-1.

How long does the Fourier GR-1 battery last?

Legacy official documentation publishes approximately 60 minutes of endurance and approximately 45 minutes of walking endurance. Because Fourier documents list different battery capacities across revisions, confirm runtime on the exact battery supplied.

How long does the GR-1 take to charge?

Legacy official documentation publishes a charging time of 315 minutes. Confirm this for the delivered battery revision rather than assuming it applies to every GR-1.

Can the GR-1 battery be hot-swapped?

Do not assume so. The public GR-1 documents reviewed here do not establish hot-swapping as a universal GR-1 capability. Ask for the correct shutdown and battery-replacement procedure.

Can the GR-1 climb stairs?

A humanoid form can support stair research, but the public specification reviewed does not provide one universal guaranteed stair geometry. Test the exact staircase and control stack.

Is the Fourier GR-1 waterproof?

No public GR-1 specification reviewed here establishes a waterproof or weatherproof IP rating. Do not use it in rain or wet environments without written confirmation for the exact configuration.

Is the GR-1 safe around people?

It should not be assumed safe for unrestricted proximity. Its mass, joint torque and fall potential require controlled operation and a task-specific risk assessment.

Is the Fourier GR-1 autonomous?

It can support autonomous research and learned policies, but it is not a turnkey general-purpose autonomous worker. Autonomy depends on the complete software, perception, data and safety stack.

Is the GR-1 discontinued?

Fourier now has newer GR-2 and GR-3 humanoids, but GR-1 documentation, model files and related resources remain publicly available, and GR-1 appears in current 2026 competition programmes. Rather than assuming either “current” or “discontinued,” buyers should ask Fourier to confirm new-unit availability, production status, support period and spare-parts coverage in writing.

Is the GR-1 still worth buying in 2026?

Potentially, yes—especially for teams with existing GR-1 research, policies or software. For a greenfield project, compare the GR-2 and other current humanoid platforms before committing.

Where can I compare the GR-1 with other humanoids?

Use the Anton Robots comparison tool and browse the humanoid robot category.

Final Verdict

The Fourier Intelligence GR-1 is no longer interesting because it was one of the early commercially developed full-size humanoids. It is interesting because it became a real research embodiment: a human-scale robot with meaningful joint torque, articulated arms, multiple end-effector configurations, secondary-development support and enough software openness to be used by external AI and robotics teams.

Its biggest weakness in 2026 is not a single mechanical specification. It is configuration certainty.

A buyer must establish exactly which GR-1 is being offered, which hand it has, which battery revision it uses, which firmware is installed, which Aurora release supports it and how long Fourier or the supplier will support the hardware.

For an experienced robotics team with a defined GR-1-specific reason to buy, that can still be a compelling platform.

For a new buyer who simply wants “a Fourier humanoid,” the GR-1 should not be selected until the newer GR-2 has been compared directly.

Bottom line: the GR-1 remains a credible humanoid research platform, but in 2026 it should be bought as a verified hardware-and-software configuration—not as a model name.

Review Methodology and Sources

This review separates four types of evidence: manufacturer specifications, manufacturer development documentation, current open-source software resources and independent research. When Fourier sources publish different specifications, the difference is disclosed rather than silently selecting the most convenient number.

• Fourier — GR-1 official legacy developer introduction, hardware architecture, compute and specifications.
• Fourier — GR-1 family product brochure covering variants, joint configuration, performance and development support.
• Fourier — GR-1 handheld-controller and standing-operation guidance.
• Fourier — GR-1 Aurora SDK quick-start documentation.
• Fourier — Aurora Python client documentation.
• Fourier — current Aurora SDK repository and supported-robot list.
• Fourier — official GRx robot model repository, GR-1 variant mapping, URDF and simulation resources.
• Fourier — official smart-actuator SDK.
• Fourier — official teleoperation repository.
• Fourier — current GR-2 product information used to establish newer-generation context.
• Fourier — current developer documentation for GR-3 and current GR-series context.
• NVIDIA Research — Isaac GR00T N1 and real-world humanoid manipulation research involving Fourier GR-1.
• GR00T N1 paper — open humanoid foundation-model research.
• HOMIE — humanoid loco-manipulation and teleoperation research using Fourier GR-1.
• Smooth Humanoid Locomotion — research resources covering GR1T1 and GR1T2 hardware.
• RoboCup Humanoid Robot Program — 2026 competition-version GR-1 programme pricing and conditions.

Anton Robots is an independent robotics marketplace. Product specifications, software support, pricing, availability and commercial terms can change. Request written confirmation for the exact configuration before purchase.