Pollen Robotics Reachy 2 Review 2026: Price, Specs & AI

Pollen Robotics Reachy 2: Quick Buyer Verdict

Reachy 2 should be evaluated as a human-scale robotics research platform, not as a finished general-purpose humanoid employee. Its strongest proposition is giving researchers a ready-made physical system for manipulation, perception, teleoperation and embodied-AI experimentation while keeping much of the software stack accessible and modifiable.

Pros

• Purpose-built for embodied AI, manipulation and human-robot interaction research.
• Two human-like 7-DoF arms with approximately 3 kg payload per arm.
• Strong perception stack with stereo RGB, depth sensing and a fixed torso RGB-D camera.
• Native VR teleoperation can generate human demonstrations for robot-learning workflows.
• ROS 2, Python SDK and open-source software provide substantial development freedom.
• Current Hugging Face LeRobot support connects physical data collection to modern imitation-learning workflows.
• Gazebo and MuJoCo options allow development before or alongside access to physical hardware.
• Omnidirectional wheeled mobility avoids the substantial complexity of bipedal locomotion research.

Cons

• Hugging Face publicly quoted US$70,000 in 2025; buyers should request a current 2026 configuration-specific quote.
• Approximately 3 kg arm payload limits heavy manipulation.
• The onboard computer is CPU-based; advanced AI inference generally requires customer-provided external compute.
• No legs means no stair climbing or human-style locomotion research.
• Pollen’s safety guidance requires close supervision and warns that arm movements do not have complete automatic collision protection.
• The mobile-base LiDAR cannot detect every obstacle around the full robot volume.
• VR teleoperation and high-bandwidth data collection depend heavily on networking and workstation setup.
• It is a research platform that still expects engineering capability from the buyer.

Our recommendation: shortlist Reachy 2 when your objective is to develop embodied intelligence, manipulation policies, teleoperation, HRI or service-robot behaviours and you value an open platform more than turnkey commercial autonomy. Define the exact experiments and compute pipeline first, then match the hardware configuration to them. Review the Pollen Robotics Reachy 2 listing before requesting a current quote.

How Much Does Pollen Robotics Reachy 2 Cost in 2026?

Hugging Face publicly stated in April 2025 that Reachy 2 could be ordered for US$70,000. We could not find a newer official September 2026 public price page that clearly supersedes that figure, so US$70,000 is best treated as a public pricing reference rather than a guaranteed current quotation.

That distinction matters because Reachy 2 exists as a modular robotics platform. Pollen Robotics has documented different configurations, including single- and dual-arm hardware and the mobile base, while a research deployment may also require external GPU compute, VR equipment, networking, workstations, custom grippers and engineering time.

Do not compare only robot prices

A US$70,000 research platform can be better value than a US$20,000 robot if it saves a research team twelve months of hardware development. Conversely, Reachy 2 is poor value if a smaller arm, mobile manipulator or lower-cost humanoid already provides the sensing and motion required for the experiment.

For a broader pricing benchmark, see our guide to humanoid robot prices in 2026.

What Is Pollen Robotics Reachy 2?

Reachy 2 is an open-source, human-scale robotic platform developed by Pollen Robotics, which became part of Hugging Face in 2025. Pollen positions the robot around research, education, embodied AI, teleoperation and real-world manipulation.

It has a humanoid upper body with two arms and an expressive head, mounted on an omnidirectional wheeled base rather than legs. That design choice is important: Reachy 2 gives researchers human-like manipulation geometry without forcing every experiment to solve the much harder problem of dynamic bipedal locomotion.

The current architecture includes two 7-degree-of-freedom arms, a 3-DoF neck, stereo and depth vision, microphones, speakers, motorised antennas, a LiDAR-equipped mobile base and a computer running the robot’s software stack.

What Reachy 2 is

• An open robotic platform for embodied-AI research.
• A bimanual mobile manipulator with human-like upper-body geometry.
• A VR-teleoperable system for collecting human demonstrations.
• A ROS 2 and Python development platform.
• A physical testbed for perception, manipulation, HRI and robot-learning research.
• A platform that can be connected to external AI models and compute infrastructure.

What Reachy 2 is not

• It is not a bipedal humanoid.
• It is not a consumer household robot.
• It is not supplied with general-purpose intelligence capable of autonomously completing arbitrary tasks.
• It is not a heavy-payload industrial manipulator.
• It is not designed for stairs, rough terrain or outdoor industrial environments.
• It should not be treated as a safety-certified collaborative worker simply because it has human-like proportions.

If you are still comparing the broader market rather than this specific research architecture, explore humanoid robots and our guide to the best humanoid robots.

Pollen Robotics Reachy 2 Specifications

The figures below combine Pollen Robotics’ current documentation with its published Reachy 2 dual-arm mobile-base datasheet. Configuration can matter, so buyers should require the specification sheet for the exact robot offered in their quotation.

Specification context matters

The 80 kg mobile-base payload is not the robot’s manipulation payload. It describes what the base can mechanically support. Reachy’s published manipulation limit remains approximately 3 kg per arm, and the safety documentation specifically warns operators not to lift more than 3 kg.

Similarly, the published eight-hour mobile-base battery figure should not automatically be interpreted as eight hours of uninterrupted dual-arm AI manipulation under every workload. Cameras, computers, networking, motion patterns and external peripherals affect the complete operating system. Validate duty cycle using the exact experiment you intend to run.

Reachy 2 Arms, Grippers and Manipulation

Manipulation is one of Reachy 2’s strongest reasons to exist. Each arm provides seven degrees of freedom using Pollen Robotics’ Orbita parallel-joint architecture: Orbita 2D mechanisms at the shoulder and elbow and an Orbita 3D mechanism at the wrist.

That gives the robot a human-like kinematic structure with significant wrist orientation freedom, while internal cable routing and the mirrored arm architecture make the platform attractive for research where the physical system itself may need to be modified.

What can Reachy 2 manipulate?

A roughly 3 kg payload per arm is suitable for many laboratory and daily-object tasks:

• Cups, bottles, packaged goods and food items.
• Small tools and research props.
• Objects used in pick-and-place and imitation-learning datasets.
• Doors, drawers or switches where forces remain within validated limits.
• Bimanual objects that require coordinated left- and right-arm movement.

It is not a substitute for a high-payload cobot or industrial arm. Long reach and awkward arm posture also reduce usable payload in practice, while applying excessive force can destabilise the complete robot.

Standard parallel grippers

The standard Reachy 2 gripper uses a Dynamixel actuator with current-based position control. Pollen publishes a maximum opening of approximately 100 mm, nominal grip force of 10 N and peak force of 50 N.

Interchangeable finger tips allow researchers to modify contact geometry and materials. More importantly, Pollen’s hardware architecture allows alternative end-effectors to be integrated; its own documentation specifically references other gripper and dexterous-hand options.

Why this matters for embodied AI

Robot-learning experiments often fail because the physical system is too inconsistent. A learning policy may be excellent in simulation but perform poorly if the real gripper slips, wrist geometry differs, camera calibration changes or actuator behaviour is not repeatable.

Reachy 2 gives researchers a relatively standardised physical platform on which perception, demonstration collection, control and policy evaluation can be repeated. That standardisation may ultimately be more valuable than raw lifting capacity.

Vision, Audio and Perception

Reachy 2 contains multiple perception systems because teleoperation, manipulation and autonomous AI each need different views of the environment.

Head vision

The head contains two RGB cameras used for stereoscopic vision and immersive teleoperation. Pollen’s technical datasheet specifies IMX296 global-shutter cameras, while current documentation describes wide-field cameras combined with an OAK-based Time-of-Flight module.

The ToF sensor supplies depth information for environmental perception and 3D mapping. Pollen’s datasheet publishes a depth range of approximately 0.20 to 5 metres, with up to 640 × 480 depth output at 45 fps.

Torso RGB-D camera

A separate fixed RGB-D camera sits in the torso and looks into the robot’s manipulation workspace. This is strategically useful: head motion can change dramatically during interaction, while a fixed body camera gives learning and manipulation systems a more consistent reference view of the arms and objects.

Audio

Two microphones are located in Reachy’s motorised antennas and a speaker system sits in the torso. Pollen designed this for both teleoperation and AI applications.

Potential research uses include:

• Speech recognition.
• Text-to-speech interaction.
• Voice-command experiments.
• Spatialised audio for remote operators.
• Multimodal human-robot interaction.

Expressive interaction

Reachy 2’s 3-DoF neck and motorised antennas are not merely cosmetic. Head orientation and small expressive movements can give human participants valuable cues about attention and intent.

That makes Reachy 2 especially relevant for HRI researchers who need a robot that can manipulate physical objects while remaining socially legible to nearby people.

Reachy 2 Mobile Base and Navigation

Reachy 2 uses an omnidirectional mobile base rather than legs. Three wheels allow the robot to translate and rotate without the large turning movements required by a conventional differential-drive base.

For manipulation research, this can be an advantage rather than a compromise. Researchers gain mobile reach across a room without needing to solve balance, foot placement, fall recovery and other bipedal problems before experimenting with the hands.

Published mobile-base specifications

Pollen documents a cylindrical base approximately 50 cm in diameter and 25 cm high, weighing 25 kg. It uses three omnidirectional wheels, wheel sensors, an IMU and an RPLIDAR S2.

The current hardware datasheet describes the LiDAR as providing up to a 30 m radius, while the SDK exposes mobile-base position, odometry, velocity and LiDAR information.

LiDAR safety layer

The SDK includes a LiDAR-based collision safety mechanism for base movement. When an obstacle enters a warning zone the robot can slow down, and movement can be blocked when an obstacle becomes too close.

This is useful, but it must not be confused with complete whole-body collision avoidance.

Pollen’s own safety documentation warns that the LiDAR observes a two-dimensional plane near the base. A table edge, human arm or other elevated obstacle may therefore sit outside that scanning plane even though Reachy’s torso or arms could hit it.

Where the mobile base works well

• Research laboratories.
• Offices and classrooms.
• Mock homes.
• Human-robot interaction spaces.
• Flat indoor service environments.

Where it does not

Reachy 2 is not designed for stairs, rough outdoor terrain or dynamic balancing. Pollen instructs users to operate it on flat, stable surfaces.

If your research question is primarily about walking, locomotion, balance or fall recovery, a biped such as the Unitree G1 is a fundamentally different—and usually more relevant—platform.

Reachy 2 VR Teleoperation

Teleoperation is one of Reachy 2’s most differentiated capabilities. Pollen Robotics provides a VR application that can place an operator effectively inside the robot: the operator sees through Reachy’s stereo cameras, controls the head and arms through natural motion and can drive the mobile base.

This matters for much more than remote control.

Teleoperation as a data-collection tool

Modern robot learning depends heavily on demonstrations. Instead of manually coding every trajectory, a researcher can perform a task through Reachy and record the robot state, images and actions.

Those trajectories can then become training data for imitation-learning policies.

Typical workflow:

1. Define a constrained task.
2. Teleoperate Reachy through successful examples.
3. Record robot states, actions and visual observations.
4. Train a policy from the demonstrations.
5. Evaluate autonomous rollouts.
6. Collect additional examples where the model fails.

This human-demonstration loop is one reason Reachy 2 fits the current embodied-AI ecosystem particularly well.

VR hardware

Pollen documentation lists Meta Quest 2 and Quest 3 among the tested/recommended options and also references HTC Vive and Valve Index compatibility.

The teleoperation application uses a computer and network link to Reachy. Current guidance emphasises reliable networking, and wired Ethernet is strongly recommended in data-collection workflows where camera streams and control traffic must remain stable.

Network quality matters

A teleoperation setup is only as good as its latency and video pipeline. Delayed stereoscopic video or inconsistent network timing can reduce operator precision and corrupt the quality of collected demonstrations.

For serious research, test:

• End-to-end motion latency.
• Video latency and dropped frames.
• Stereo-camera synchronisation.
• Network bandwidth during recording.
• Emergency-stop access.
• Behaviour after connection loss.

ROS 2, Python SDK and Open-Source Software

Reachy 2’s software stack is one of its biggest advantages over robotics platforms that expose only a limited vendor API.

Pollen’s current architecture is built around ROS 2. A Python SDK communicates with the robot through gRPC services, while the underlying server bridges those commands into the ROS 2 hardware and control environment.

Python SDK

The Reachy 2 SDK lets developers interact with:

• Arms and grippers.
• Head and antennas.
• Mobile base.
• Cameras.
• Audio.
• Robot state and motion commands.
• Inverse and forward kinematics.

For many AI researchers, this high-level Python access substantially reduces the robotics-specific work needed before running an experiment.

ROS 2

Teams that need deeper integration can operate closer to the ROS 2 layer. This is relevant for navigation, planning, custom hardware, robot state, visualisation and integrating Reachy with an existing robotics research stack.

Pollen’s technical documentation states that the software exposes standard ROS 2 interfaces, including robot states and transforms, while providing kinematic services.

Open-source advantage

Open source does not mean that every hardware component is free or trivial to reproduce. Its value is that researchers can inspect, extend and integrate substantial parts of the software stack instead of depending entirely on black-box behaviour.

That matters when the objective is not simply to use a robot but to study robotics itself.

Pollen also publishes CAD resources and maintains public repositories around the SDK, APIs, simulation, teleoperation and related tools.

Reachy 2, LeRobot and Embodied AI

The Hugging Face connection is strategically important.

Hugging Face acquired Pollen Robotics in April 2025 after developing LeRobot, its open-source robotics framework for datasets, models and robot-learning workflows. Reachy 2 now sits at the intersection of that software ecosystem and a commercially available human-scale robot.

Current LeRobot documentation includes dedicated Reachy 2 support for recording demonstrations, replaying episodes, training policies and evaluating them on the physical robot.

A modern Reachy 2 learning loop

A research team can now structure work roughly as:

teleoperate → record → dataset → train → evaluate → collect failures → retrain.

That is much closer to the workflow used in current physical-AI research than the traditional approach of manually programming every robot behaviour.

What LeRobot does not mean

LeRobot support does not mean Reachy 2 arrives with general intelligence.

The robot still requires:

• A task definition.
• Demonstration or training data.
• An appropriate model or policy.
• Compute for training and often inference.
• Evaluation.
• Safety supervision.
• Engineering around failure cases.

A policy trained to pick up a cup does not automatically know how to load a dishwasher, clean a table or operate in an unseen kitchen.

External AI compute

Reachy’s internal SolidRun computer is described by Pollen as a CPU-based industrial system. The manufacturer’s hardware documentation explicitly notes that customer-provided computers are used for AI processing.

This is an important procurement point.

If your research involves large vision models, diffusion policies, VLAs or other compute-intensive inference, include a suitable GPU system in the project architecture and budget.

Reachy 2 Simulation: Gazebo and MuJoCo

Pollen provides simulation workflows so developers can interact with a virtual Reachy 2 before running code on the physical machine.

Current documentation supports both Gazebo and MuJoCo through a Docker-based environment.

Why simulation matters

Simulation lets teams:

• Learn the SDK before hardware arrives.
• Prototype trajectories without risking the physical robot.
• Develop software remotely.
• Test kinematics and application logic.
• Build CI workflows around robot code.
• Experiment with synthetic tasks and scenes.

Pollen’s simulator is designed so much of the same SDK interface used against the virtual robot can later be pointed at the physical robot.

Do not assume perfect sim-to-real transfer

A digital model cannot perfectly replicate gripper friction, cable behaviour, camera noise, actuator variation, object compliance or real-world contact.

For manipulation research, simulation should shorten development—not eliminate physical validation.

Policies trained entirely in simulation still need systematic real-world evaluation before anyone should rely on them.

Reachy 2 Battery Life and Operating Time

Pollen publishes up to approximately eight hours of battery life for the mobile base. The base uses a 24 V, 35 Ah LiFePO₄ battery with a battery-management system.

The published battery specification is encouraging for a research platform, but buyers should separate mobile-base battery capacity from guaranteed whole-system experimental runtime.

Arm motion, mobile-base movement, onboard electronics and connected peripherals all affect energy use. External GPU systems used for AI typically have their own power requirements and are not represented by the robot battery figure.

Measure your actual experiment

Before designing all-day data collection, test:

• Runtime during continuous bimanual movement.
• Runtime during mobile manipulation.
• Battery behaviour during VR teleoperation.
• Time required for charging.
• Whether your experimental protocol can tolerate interruptions.
• Whether external compute and networking remain independently powered.

For university labs, repeatability may matter more than absolute maximum runtime. A reliable four-hour experiment is more useful than an eight-hour theoretical specification that changes dramatically with workload.

Reachy 2 Safety and Deployment Limitations

This is one of the most important sections for anyone interpreting Reachy 2 demonstrations.

Pollen’s current safety documentation is explicit: operators must remain vigilant, be ready to use the emergency stop and supervise the robot’s surroundings.

Arm collision awareness is limited

Pollen warns that Reachy can follow commanded motions without automatically detecting all obstacles, people or self-collisions.

That means a researcher cannot treat the robot like a fully safety-rated collaborative industrial manipulator.

Pollen recommends a safety area around the robot and advises users to remain outside the rear area where perception is particularly limited.

Mobile-base LiDAR is not whole-body perception

The mobile base includes anti-collision behaviour based on its LiDAR, but the sensor observes obstacles at a specific height.

A desk, shelf, person or object above that scanning plane may remain a collision risk for the arms or torso.

Payload safety

Pollen instructs users not to exceed approximately 3 kg while lifting and recommends keeping loads close enough to the torso to reduce tipping risk.

Sharp, heavy or otherwise hazardous objects require additional precautions.

Flat floors only

The robot should operate on stable, level ground. It is not a rough-terrain platform.

Human-robot experiments need protocol design

For HRI research, define:

• Who may enter the robot workspace.
• Where the emergency stop is located.
• Maximum permitted velocity and force.
• Approved objects and tools.
• What happens after network loss.
• Who is responsible for supervising autonomous policies.

A learned policy should never be assumed safe merely because it succeeded in previous episodes.

Reachy 2 Research Adoption and Real-World Evidence

Reachy 2 does not yet have the kind of production-factory ROI case library associated with mature industrial robots. That is not the right evidence standard for this product.

The relevant evidence is whether serious research organisations are using the platform and whether the surrounding development ecosystem is active.

Hugging Face states that Reachy 2 is already used in laboratories including Cornell University and Carnegie Mellon University. Its acquisition announcement also cites adoption of the broader Reachy platform by organisations including Accenture, CEA and CNRS.

Pollen Robotics also has a longer history in telepresence and manipulation. A previous Reachy system placed second in the ANA Avatar XPRIZE in 2022, demonstrating the lineage behind Reachy 2’s emphasis on immersive teleoperation and physical interaction.

What this proves

It supports several conclusions:

• Reachy is more than a one-off concept robot.
• Researchers outside Pollen Robotics have access to physical systems.
• The company has years of experience building teleoperated manipulation platforms.
• The software ecosystem is being actively developed.

What it does not prove

It does not prove that Reachy 2 can independently perform every household or service task seen in demonstration videos.

Research demonstrations often combine teleoperation, scripted control, learned policies, controlled environments and task-specific engineering. Buyers should ask exactly which operating mode produced any capability they want to reproduce.

Best Pollen Robotics Reachy 2 Use Cases

1. Embodied AI research

Best overall use case. Reachy 2 provides a physical platform where AI models must deal with perception, geometry, contact, uncertainty and real objects rather than text or simulated pixels alone.

It is particularly relevant to teams researching vision-language-action models, imitation learning, diffusion policies, reinforcement learning and multimodal robot control.

2. Imitation learning and demonstration collection

The VR teleoperation workflow lets researchers generate real robot demonstrations using human movements.

This can dramatically simplify dataset collection for tasks such as reaching, grasping, sorting, handovers and bimanual manipulation.

3. Bimanual manipulation

Two 7-DoF arms allow experiments that cannot be reproduced with a single robotic arm.

Examples include holding an object with one hand while manipulating it with the other, handovers, container opening, coordinated placement and human-style two-handed interaction.

4. Human-robot interaction

Reachy’s expressive head, microphones, speakers, cameras and antennas make it useful for studying gaze, attention, social behaviour, communication and physical interaction.

It sits between a purely social robot and a conventional mobile manipulator because it can interact with people while also manipulating real objects.

5. Telepresence and remote manipulation research

VR teleoperation makes Reachy relevant where researchers need to study remote physical presence rather than only video conferencing.

The platform’s history in avatar robotics reinforces this use case.

6. Service-robot prototyping

Reachy 2 can act as a development platform for future service applications in hospitality, retail, laboratories, offices or assisted environments.

The key word is prototyping. A successful Reachy experiment does not automatically constitute a commercially deployable service robot.

If commercial service deployment is the real objective rather than research, compare dedicated service robots as well.

7. Robotics education

Advanced universities can use Reachy to teach kinematics, ROS 2, perception, manipulation, teleoperation, robot learning and human-robot interaction on one integrated system.

For simpler teaching objectives or larger student cohorts, however, considerably cheaper educational platforms may provide better economics. Explore educational robots before assuming every robotics course needs a US$70,000-class platform.

8. AI model evaluation in the physical world

Reachy 2 can become a repeatable embodiment on which researchers compare models.

Instead of asking whether two policies perform well on different robots, a laboratory can evaluate them against the same cameras, kinematics, grippers and environment.

That makes Reachy useful as experimental infrastructure, not merely as a robot demo platform.

When Pollen Robotics Reachy 2 Is Not the Right Robot

Reachy 2 is a strong platform, but several requirements should immediately push buyers toward alternatives.

• Bipedal locomotion research: Reachy uses a wheeled base; choose a legged humanoid if walking and balance are central to the experiment.
• Heavy industrial manipulation: approximately 3 kg per arm is far below the payload of many cobots and industrial arms.
• Turnkey factory automation: Reachy is a research and development platform, not a pre-engineered production cell.
• Outdoor or rough terrain: the robot is designed around flat indoor surfaces.
• Autonomous household labour today: research demonstrations should not be interpreted as a commercially ready home-assistant capability.
• Minimal technical staffing: buyers should expect Python, robotics, ML or systems-integration work.
• Safety-critical unsupervised operation: current manufacturer guidance requires active awareness and supervision.
• Lowest-cost manipulation research: a pair of smaller arms or a tabletop system may answer the same scientific question for much less.

The most expensive mistake would be buying Reachy because your team “wants a humanoid” before identifying why human-scale bimanual embodiment is actually required.

Pollen Robotics Reachy 2 vs Unitree G1, PAL Robotics TIAGo Pro and Other Research Robots

Reachy 2 sits in an unusual position because it combines humanoid manipulation with wheeled mobility. Some competitors are bipedal humanoids; others are research mobile manipulators.

The correct alternative therefore depends on the research question.

Which one should you choose?

• Choose Reachy 2 when manipulation, teleoperation, robot-learning datasets and embodied AI are the priority.
• Choose Unitree G1 EDU when dynamic bipedal locomotion and whole-body learning are central to the research programme.
• Choose TIAGo Pro when you want a more conventional mobile-manipulation research architecture with torque-aware arms and navigation emphasis.
• Choose NAO when HRI and education matter more than adult-scale manipulation.

These platforms should not be compared using DoF or purchase price alone. Ask which physical embodiment makes the scientific problem easier to study.

Use the Anton Robots comparison tool to narrow available platforms.

Is Pollen Robotics Reachy 2 Worth It?

Reachy 2 is worth it when the value lies in accelerating robotics research rather than replacing paid labour immediately.

For a university or AI lab, the financial question is different from evaluating an industrial cleaning robot or AMR.

The alternative to buying Reachy may be:

• Designing a custom dual-arm robot.
• Integrating actuators and motor controllers.
• Building perception hardware.
• Developing kinematics and ROS interfaces.
• Building teleoperation from scratch.
• Creating simulation assets.
• Maintaining all of that hardware internally.

If Reachy eliminates a year of engineering before the actual AI research can begin, the acquisition cost may be easy to justify.

Build the value model around research output

A useful model is:

Research value = engineering time avoided + experiments enabled + datasets produced + researcher productivity + grant/project value − total platform cost.

Then evaluate whether a cheaper platform can deliver the same outputs.

Costs to include

• Robot and configuration.
• Shipping and import costs.
• VR equipment.
• External GPU compute.
• Data storage.
• Replacement components.
• Integration and researcher time.
• Maintenance and support.

Benefits to quantify

• Months of hardware development avoided.
• Number of researchers able to share one standardised platform.
• Experiments possible per semester or project.
• Ease of reproducing external research.
• Quality and scale of demonstration datasets.
• Ability to move a policy from simulation to a physical human-scale robot.

A practical go/no-go test

Before buying, write down three experiments you intend to run during the first six months.

For each experiment, specify:

input → robot behaviour → data collected → model or hypothesis tested → measurable research output.

If that exercise is difficult, the lab may be buying an impressive robot before defining the research programme.

Pollen Robotics Reachy 2 Buying Checklist

1. Define the research objective. Is the programme about manipulation, teleoperation, HRI, learning, navigation or whole-body robotics?
2. Confirm Reachy’s embodiment is necessary. Determine why two human-scale arms and mobile reach are needed.
3. Confirm the exact configuration. Specify number of arms, grippers, mobile base, cameras and included accessories.
4. Confirm the current price. Treat the publicly announced US$70,000 figure as a reference until a current written quote is received.
5. Map the compute architecture. Choose GPU workstation, server or cloud infrastructure for AI workloads.
6. Define the data pipeline. Decide what observations and actions will be recorded, where datasets will live and how they will be versioned.
7. Plan teleoperation. Select VR hardware, workstation and network architecture.
8. Validate networking. Test required camera bandwidth, latency and Ethernet topology.
9. Choose end-effectors. Confirm whether the standard parallel grippers can perform the target tasks.
10. Review safety. Define workspace, supervision, emergency-stop access and permitted autonomous operation.
11. Test simulation first. Prototype with the public simulation environment where practical.
12. Define acceptance tests. Include arm reach, payload, grasp success, teleoperation latency, camera quality and data-recording reliability.
13. Review support. Ask about warranty, spare parts, response times, repairs and international service.
14. Plan the first six months. Have experiments ready before the physical robot arrives.

Pro tip: do not start the procurement document with “We need a humanoid robot.” Start with “We need to collect and evaluate X type of physical interaction data.” The correct embodiment usually becomes much clearer after that.

How to Buy Pollen Robotics Reachy 2

Reachy 2 is a specialised research purchase rather than an ordinary ecommerce robot.

A useful enquiry should explain the intended research programme, manipulation requirements, preferred configuration, AI stack, end-effectors, location and desired delivery date.

Before requesting a quote, prepare:

• Institution or company name.
• Research or development objective.
• Required robot configuration.
• Target manipulation tasks.
• Payload requirements.
• Preferred gripper or hand architecture.
• VR and teleoperation requirements.
• Existing ROS 2 and AI infrastructure.
• Delivery country.
• Target project start date.

Review the Pollen Robotics Reachy 2 product page and the Pollen Robotics manufacturer profile, then contact Anton Robots if you need help comparing the platform with alternatives.

If the embodiment is still unclear, use Find My Robot before committing to a specific platform.

What Matters for Reachy 2 in 2026?

Deeper Hugging Face and LeRobot integration

The most important change in Reachy 2’s position is not a new motor or cosmetic redesign. It is the expanding connection between Pollen Robotics hardware and Hugging Face’s open robotics ecosystem.

Current LeRobot documentation now includes an explicit Reachy 2 workflow for recording robot data, replaying episodes, training policies and evaluating them.

For embodied-AI teams, this materially strengthens Reachy’s proposition as a standard research platform.

Current VR-to-dataset workflows

Reachy 2 teleoperation is increasingly useful as the front end of a learning pipeline rather than simply as a remote-control demo.

Current LeRobot documentation combines Reachy teleoperation with dataset recording, meaning human demonstrations can feed directly into policy-training workflows.

This is strategically important because high-quality real-world data is becoming one of the main constraints in physical AI.

MuJoCo development

Pollen’s simulation environment now includes MuJoCo alongside Gazebo, giving researchers another route for prototyping and integrating modern learning workflows.

Simulation remains an evolving part of the platform and should not be assumed to reproduce every sensor or contact effect of the physical robot.

Pollen Robotics is now part of Hugging Face

Hugging Face acquired Pollen Robotics in April 2025.

For buyers, this changes the context around the platform. Reachy 2 is no longer simply a robot from a small independent hardware company; it now sits beside Hugging Face’s models, datasets, LeRobot library and broader open-source AI ecosystem.

That does not guarantee future capabilities, but it is a meaningful factor for research teams deciding which platform is most likely to remain relevant to open embodied-AI development.

Pollen Robotics Reachy 2 FAQ

How much does Reachy 2 cost?

Hugging Face publicly stated in April 2025 that Reachy 2 could be ordered for US$70,000. We could not identify a newer official public 2026 list price that clearly replaces that figure, so buyers should request a current quote and confirm exactly which hardware configuration is included.

Is Reachy 2 open source?

Yes. Pollen Robotics develops Reachy around an open-source hardware and software philosophy, with public SDK, API, teleoperation and other repositories. The Reachy 2 Python SDK is published under the Apache 2.0 licence.

Is Reachy 2 a humanoid robot?

Yes, but with an important qualification. Reachy 2 has a human-like torso, head and two articulated arms, but it moves on an omnidirectional wheeled base instead of legs. It is therefore better described as a wheeled humanoid or human-scale mobile manipulator than as a biped.

Can Reachy 2 walk?

No. Reachy 2 uses a three-wheel omnidirectional mobile base. It cannot walk or climb stairs like a bipedal humanoid.

How many degrees of freedom does Reachy 2 have?

Each arm has seven degrees of freedom and the neck has three. Additional actuation exists in the grippers, antennas and mobile base, so total actuator counts depend on how DoF are defined.

How much can Reachy 2 lift?

Pollen publishes an arm payload of approximately 3 kg per arm and its safety guidance tells operators not to exceed that limit. Practical payload also depends on reach and object position.

How much can the Reachy 2 gripper hold?

The standard parallel gripper has a published nominal force of 10 N and peak force of 50 N, with approximately 100 mm maximum opening. Grasp performance still depends heavily on object geometry, friction and finger tips.

Does Reachy 2 have hands?

The standard configuration uses parallel grippers rather than five-finger human hands. Pollen’s architecture allows alternative end-effectors to be integrated, including research-oriented dexterous-hand options.

Does Reachy 2 use ROS?

Yes. Reachy 2’s current core software is based on ROS 2, with a higher-level Python SDK available for easier robot programming.

Does Reachy 2 support Python?

Yes. Pollen provides a dedicated Reachy 2 Python SDK with APIs for robot movement, cameras, audio, mobile base and other functionality.

Does Reachy 2 work with Hugging Face LeRobot?

Yes. Current LeRobot documentation includes dedicated Reachy 2 support for data recording, replay, training and evaluation.

Can Reachy 2 collect data for imitation learning?

Yes. VR teleoperation can be used to produce human demonstrations, and current LeRobot workflows can record Reachy 2 observations and actions into datasets for model training.

Can Reachy 2 run AI models onboard?

Some processing can run locally, but Pollen describes the integrated SolidRun Bedrock v3000 as a CPU-based computer and states that customer-provided hardware is used for AI processing. Compute-intensive vision or robot-learning models will generally need external GPU resources.

What cameras does Reachy 2 have?

The head includes stereo RGB cameras and Time-of-Flight depth sensing. A separate fixed RGB-D camera in the torso observes the manipulation workspace.

Does Reachy 2 have LiDAR?

Yes. The omnidirectional mobile base includes an RPLIDAR S2 used for environmental sensing and base collision-safety functions.

Is Reachy 2 autonomous?

It can run autonomous software and learned policies, but it should not be described as a ready-made autonomous general-purpose robot. Researchers must create, integrate or train the behaviours required for their application.

Can Reachy 2 do household chores?

Reachy 2 has been demonstrated performing household-style manipulation tasks under research conditions using teleoperation and learned behaviours. That is evidence of the platform’s research potential, not evidence that a purchased robot can independently operate a home out of the box.

Can Reachy 2 be teleoperated?

Yes. VR teleoperation is a core capability. Operators can control Reachy’s head and arms, see through its stereo cameras and move the mobile base.

Which VR headsets work with Reachy 2?

Pollen documentation lists Meta Quest 2 and Quest 3 among the tested or recommended devices and also references HTC Vive and Valve Index support. Buyers should verify current software compatibility before purchasing VR equipment.

How long does Reachy 2’s battery last?

Pollen publishes up to approximately eight hours for the mobile base. The actual complete-system operating time depends on robot motion, electronics, experimental workload and attached equipment.

Can Reachy 2 work outdoors?

Reachy 2 is primarily an indoor research platform. Pollen’s safety guidance specifies flat, stable operating surfaces, and public documentation does not position the robot as a weatherproof outdoor system.

Is Reachy 2 safe around people?

It is designed for human-robot research, but that does not mean unsupervised operation is inherently safe. Pollen requires active supervision, access to an emergency stop and careful workspace management, and warns of important collision-detection limitations.

Can Reachy 2 climb stairs?

No. The robot uses a wheeled mobile base and is intended for flat surfaces.

Can Reachy 2 be simulated?

Yes. Pollen provides Docker-based simulation workflows for Gazebo and MuJoCo.

Is Reachy 2 good for universities?

Yes, particularly for advanced robotics and AI laboratories studying manipulation, embodied AI, imitation learning, teleoperation and HRI. It may be unnecessarily expensive for introductory robotics teaching where smaller platforms provide the same educational outcome.

Is Reachy 2 better than Unitree G1?

They optimise for different research questions. Reachy 2 is particularly strong for bimanual manipulation, VR teleoperation and embodied-AI data collection on a stable wheeled base. Unitree G1 is more relevant when bipedal locomotion and whole-body dynamic control are central to the research programme.

Who owns Pollen Robotics?

Hugging Face acquired Pollen Robotics in April 2025. Pollen continues to develop open robotics hardware as part of the Hugging Face ecosystem.

Is Reachy 2 worth US$70,000?

It can be for a laboratory that would otherwise spend substantial engineering time building a comparable bimanual mobile manipulation platform. It is difficult to justify if the research question could be answered with a smaller robot, one arm or simulation alone.

Final Verdict: Should You Buy Pollen Robotics Reachy 2?

Buy or shortlist Reachy 2 if you need a human-scale, open and programmable physical platform for embodied AI, bimanual manipulation, teleoperation or human-robot interaction—and your team wants to spend its time researching intelligence rather than building an entire robot from scratch.

Its strongest combination is not any one specification. It is the integration of two 7-DoF arms, rich perception, expressive interaction, omnidirectional mobility, VR teleoperation, ROS 2, Python, simulation and Hugging Face LeRobot.

The limitations are equally important. Reachy 2 is wheeled rather than bipedal, arm payload is approximately 3 kg, advanced AI needs external compute and the robot still requires engineering and close safety supervision. It should not be purchased as a turnkey autonomous employee.

For the right research group, those are acceptable compromises because the objective is not to purchase finished labour—it is to purchase an extensible physical embodiment on which new robot intelligence can be built and measured.

The smartest approach is to define the first experiments before procurement: which tasks will be demonstrated, which signals will be recorded, what models will be trained and how success will be evaluated.

Ready to evaluate the platform? View Pollen Robotics Reachy 2 at Anton Robots or request help comparing Reachy 2 with other humanoid and research robots.