Research

Having complete and unfettered access to all parts of the world is important for NEO to accomplish tasks in home environments, and to enable our Redwood AI to learn from the broadest set of physical interactions possible. Our latest RL controller provides Redwood with a complete mobility toolkit to access the world, including natural walking in any direction, sitting, standing, kneeling, getting up from the floor, going up and down stairs with stereo vision. Bringing all of these capabilities together for the first time in a unified controller represents an important milestone in unlocking the full potential of humanoid robots.

SIT & STAND

Full stairs

Bridging Natural Movement with Omnidirectional steering

The past decade has seen rapid progress in legged robot locomotion, primarily driven by torque-transparent actuators, deep Reinforcement Learning (RL) algorithms, and GPU-accelerated simulation. Using off-the-shelf software packages and a desktop GPU, it is now possible to train a bipedal robot to stand upright in simulation and follow walking direction commands in under an hour. Even though the policy is trained completely in a simulation environment whose physics are merely an approximation of the real world, it is trained on so many randomized physics parameters – e.g. friction, mass, sensor noise – that the model ends up being robust to the real world’s physics parameters. Once trained, this walking controller can receive walking direction commands from either the teleoperator or the AI model (e.g. Redwood), and then translate those high level directions into the dynamic, contact-aware interactions with the world.



Beyond walking and turning, bipeds can also perform side-stepping. This is useful for navigating the close quarters of a kitchen or the space between the sofa and the coffee table, where the footprint would be too small for a wheeled base robot.



However, these basic walking RL controllers often require additional “shaping rewards” to achieve a natural human-like gait in all directions. These tend to be highly specialized to walking, which means that the same process of hand-tuning rewards must be repeated for every new behavior. Gait patterns can vary based on the direction of walking, so this often requires unique shaping terms based on the direction of motion.

Is there a more scalable way to increase controller capabilities without hand-written shaping rewards for each of them? One method is to collect motion capture references from humans moving naturally, retarget them to NEO’s joints and body, and then train the RL controller to match those kinematic reference trajectories.

Because the reference only specifies where the body should be, the RL controller still needs to figure out how to keep the robot stable, while “keeping tempo” and tracking the reference trajectory as closely as possible.

Using these techniques, it is possible to over-fit a policy to track a single human motion capture trajectory and achieve very dynamic and fluid motions such as dancing or walking. Examples of “single trajectory replay controller” are shown below for natural running and pivoting:

Pivot



These behaviors, while elegant, are not readily useful for general purpose tasks as they only can replay a single trajectory. They do not expose a steerable interface with which a high level policy like the AI model can execute the right actions. It is also not obvious how to transition smoothly from one reference to another, as motion capture datasets rarely include the transition behaviors between arbitrary task-centric motions, e.g. switching from shuffling quickly side-to-side to a skipping motion.

To handle multiple trajectories, we could train the controller to follow multiple mocap references, taking in the encoded kinematic trajectory as an input. However, this approach encounters a teleoperation UX problem: it is not obvious how one provides the high-dimensional kinematic trajectories at test-time using a more limited input device like a gamepad joystick or a VR controller. The model is trained on high-dimensional kinematic trajectories with nuanced rhythm and periodicity, but the commands provided by teleoperation are coarse-grained, which results in the RL controller interpreting it into an unnatural gait.

How does one achieve steerability and robustness, while still achieving the fluidity of motion capture-based RL training? We developed a two-stage controller, consisting of a high level “kinematic planner” to synthesize kinematic targets that resemble human motion capture data, and a low level controller that tries to achieve those plans.



The lower level RL controller takes as input a kinematic reference trajectory of body poses that it must attempt to track while maintaining balance. This is paired with a high level motion generator model that is trained with supervised learning to convert input commands like joystick direction to the richer kinematic trajectories. The generative model also plays the role of smoothing transitions during behavior changes.

An important reason for having robots with legs in the home is to traverse stairs. We develop a “stair mode” in our controller which engages the use of stereo RGB vision to infer the height of the floor around NEO.

To climb up and down stairs in a graceful way, NEO’s RL controller must anticipate the necessary height of each step well in advance of making contact with that step. Unlike most humanoid robots, which employ a time-of-flight depth sensor or a lidar to estimate the floor plane, NEO’s RL controller is purely vision based. Depth is predicted directly from the RGB stereo pair, and this is fused with the proprioceptive history for NEO to figure out how and where to step.



Stairs are not always idealized; through domain randomization in simulation, the controller is also robust enough to support side-stepping and handling stairs of mixed heights.

There are numerous home chores that require NEO to work at floor height for extended periods of time: removing a stain from the carpet, reorganizing the bottom drawer of a cabinet, packing a suitcase, and sorting socks. We extend our RL controller to be able to safely sit, kneel and lie down on the floor, as well as get back up from each of these poses.

Kneel

Sit & stand

The controller provides an “action interface” for which teleoperation or Redwood AI is able to interact in a safe, contact-rich manner with the physical world. To demonstrate the controllability of the natural walking behavior, we fine-tuned the Redwood model to do a soccer ball dribbling task.

Here is Redwood interacting with this new RL controller, where it predicts whole body joint targets and walking pelvis velocities from vision. The controller then translates those intents into the specific forces applied by the leg to walk in the direction of the ball.

We have developed the first general-purpose, fully AI and teleoperation compatible controller that unlocks the full kinematic workspace that is available to a bipedal humanoid robot. This will enable us to train Redwood AI to fully explore the entire state space of the home: every high and low shelf, every nook and cranny, every floor.

We will then use that data to make an AI like no one has ever seen.

Redwood AI: Mobility

We are excited to introduce Redwood, 1X’s breakthrough AI model that we will be deploying to homes. Redwood is a vision-language transformer tailored for the humanoid form factor and capable of performing end-to-end mobile manipulation tasks like retrieving objects for users, opening doors, and navigating around the home. Redwood empowers NEO Gamma to learn from real-world experience, on top of hardware designed for compliance, safety, and resilience.

: Handles variation in tasks—like picking up never-before-seen objects in unfamiliar locations. Trained on a large dataset of teleoperated and autonomous episodes from EVE and NEO, Redwood exhibits emergent behaviors such as choosing

 Redwood is among the first VLAs to control locomotion jointly with manipulation, enabling bracing and leaning behaviors during manipulation.

 Allows NEO to position itself precisely for tasks, perform actions that require movement across space, and manipulate objects while on the move.

Redwood is compute-efficient and runs fully on NEO’s onboard embedded GPU.

In order to power autonomy on both EVE and NEO platforms, Redwood fuses pre-trained language embeddings, vision tokens from a pre-trained vision transformer, and proprioception embeddings from a sequence of joint positions and joint applied forces. These are passed through several more transformer blocks, which extract a latent representation vector. We decode this representation into EVE or NEO actions using a 



Generalizing to manipulating new objects in locations not seen in the training data is crucial for the model to work in home environments, where the home is never in the exact same configuration twice. This is achieved by training on a diverse dataset gathered on NEOs in 1X offices as well as employee homes.

To further improve generalization to new scenarios despite its small size (160M parameters), Redwood is trained not only to predict actions, but a variety of “cognitive” prediction targets like estimating the current location of NEO’s hands and relevant objects in image space. These cognitive tasks help ground NEO’s visual representations and allow it to generalize better to unseen environments, despite having a small model size. Below, we show a continuous take of Redwood being able to grasp unseen bottles from a variety of locations.

Whole Body Control and Multi-contact manipulation

Manipulation and locomotion behaviors are typically decoupled in most robotic systems. However, manipulation in the home necessitates going beyond picking small objects on counter-tops and tables: humans use their legs, hips and spine to bend down to pick toys and clothes off the ground, and lean into heavy doors when pushing them open. These “whole body control” tasks make it impossible to cleanly separate locomotion and manipulation.

To enable similar capabilities on NEO, Redwood predicts not only the arm and hand commands, but also walking, manipulation, and pelvis pose commands simultaneously. This greatly expands the kinematic reach and payload capacity NEO can work with.



Coordinating all parts of the body to engage with the environment also enables multi-contact manipulation, such as bracing a hand against a wall when pulling open heavy doors.

Solving chores requires combining manipulation with navigating across the home. In real home tasks, the objects of interest are rarely all in front of the robot at the start. Furthermore, navigating to and getting close enough to an object to grasp it needs to take into account the way the model will choose to pick up the object. If one does not train navigation skills jointly with those for manipulation, then a separate navigation stack may fail to position the robot in an optimal position to grasp the target object. Vice versa, if manipulation behavior does not take into account navigation behavior that might follow it, this could lead to unwanted collisions or carrying the object across the room in an unsafe way.

To that end, Redwood is trained on a large diverse set of object navigation and pick-and-place demonstrations within the home and is trained to plan navigation and manipulation behaviors jointly. An emergent property of training from these demonstrations is that Redwood can automatically decide to use the left, right, or both hands to pick up an object.

Running Redwood onboard allows NEO to be deployed in more diverse environments: in basements, in the garden, in homes with spotty Internet infrastructure, in wilderness campsites.

To that end, Redwood is a 160M parameter transformer model that runs on NEO’s onboard GPU at around 5hz. To pack as much intelligence as possible into a relatively small number of parameters, we’ve found that the additional cognitive losses help with grounding the representations, especially in unseen environments.

Voice control is an intuitive interface to interact with general-purpose robots in the home. Using an offboard speech-to-speech LLM, we extract the goal the user intends to command NEO with from a conversational context, and then convert the command into a vector offboard using a sentence encoder. This vector is then passed as an input into the Redwood model, which is trained on thousands of such text embeddings.

Large-scale behavior cloning methods typically only imitate successful demonstrations. Redwood is trained to learn from both successful and failure rollouts, allowing NEO to improve from any interaction it has with the world regardless of success. The failure rollouts provide supervision signals on the cognitive prediction heads, which helps prevent overfitting to a relatively narrow distribution of states seen during successful demonstrations. The successful demonstrations supervise both the action diffusion heads and the cognitive predictions.

We think that general purpose autonomous humanoids, with their intelligence incubated in the home, will be a generation-defining technology that reshapes quality of life for the elderly, for busy parents, and for use cases that scarcely cannot be imagined today. We’re looking for driven, high-agency engineers to help us scale up Redwood to the next level, and to deploy a production-grade AI in as many homes as possible this year. If working on Redwood excites you, we have a large number of open roles in Palo Alto:

Research Engineer, Reinforcement Learning

Redwood AI

Developing AI for humanoid robots involves tackling many open research challenges – in safety, dexterity, visual understanding, and much more. It helps to compare notes with other labs tackling similar challenges, in order to accelerate progress towards a future of NEOs doing all the tasks needed to keep your home in order autonomously.

To that end, 1X AI and NVIDIA are pleased to announce our research collaboration effort. As a first step, the teams worked together to prepare an autonomy demo for Jensen Huang’s GTC 2025 Keynote, featuring NEO doing a dish loading task autonomously.

The following is a look into where, how and when we taught NEO to do the dishes with the NVIDIA team.



To make this collaboration possible, the 1X AI Team created a dataset API for NVIDIA to access data collected from 1X offices and employee homes, and an inference SDK to serve model predictions at a continuous 5Hz vision-action loop using an onboard NVIDIA GPU in NEO’s head or an offboard GPU.

A crucial step when onboarding a new learning codebase onto NEO is to verify correctness, i.e., overfitting a baseline model to a small amount of demonstration data and making sure that the time synchronization between images and actions is consistent all the way from data collection to training to runtime inference.

We demonstrate this by working with the NVIDIA GEAR team to train a single end-to-end neural network based on the 

 model to autonomously grasp a cup, hand it over to the other hand, and place it in a dishwasher to showcase how NEO fits compactly into the kitchen space while still having the kinematic reach to carry the cup from sink to dishwasher.

This is a good “first task” to learn because it checks for basic compatibility of an external research codebase with the logging and inference architecture. The obvious next step after verifying correctness is to feed thousands of hours of internally collected NEO data into the model.

Over the course of a week, our teams developed this model at a 1X employee’s home, swapped notes on action spaces, control frequencies, and other imitation learning tricks needed to get good performance on NEO Gamma. Moments like these – where friends are just hanging out in the home while a NEO does dishes in the background – will soon become an everyday occurrence.



When working in homes, the safety of NEO Gamma becomes particularly evident. NEO’s mechanically compliant and safe design allowed engineers to get in extremely close quarters with the robot while testing a variety of experimental architectures.


Our teams are both looking forward to continuing to learn from each other and push the industry forward. We hope that together we can accelerate our path to humanoids living and learning among us and providing a helping hand wherever one is needed.


 team and Jensen Huang for being gracious with their time and having us be a part of the NVIDIA experience at GTC. 

Additionally, thank you to 

 team for designing the jacket and capturing the moment.

1X & NVIDIA Research Collaboration 

We see World Models as a grand challenge in robotics. They have the potential to solve general purpose simulation and evaluation, enabling robots that are safe, reliable, and intelligent.

 work on our robot world model which imagines possible futures given action proposals. We also announced our first challenge: compression, which focuses on minimizing training loss across a diverse robot dataset. The lower the loss, the better the model understands the training data. This challenge is still active, offering a $10k prize to the first submission that achieves a loss of 8.0 on our private test set. Our Github repo provides code and pretrained 

Today, we are announcing the next phase of the World Model Challenge: Sampling.

Sampling focuses on generating realistic future outcomes in video sequences by predicting the next frame given a sequence of prior frames. The goal is to produce coherent and plausible continuations of the video, accurately reflecting the dynamics of the scene. We encourage you to explore a variety of future prediction methods beyond traditional next-logit prediction. Techniques such as Generative Adversarial Networks, Diffusion Models, and MaskGIT are all welcome for generating the next frame. To be competitive, submissions should achieve a PSNR of around 26.5 or above. We will open our evaluation server for submissions and release our metric in March 2025.

To help accelerate research in this direction, we’re 

 of 100 hours of raw robot video alongside our robot state sequences which enable world model training. Our raw videos will be shared under the 

 license, and we will continue to share tokenized datasets under 

We’re also thrilled to announce that we’re partnering with NVIDIA’s World Models team to further tokenize our video sequences with their newly-announced 

. NVIDIA’s work in visual tokenization and quantization creates highly compressed, temporal representations of our robot data, optimized for such research. The Cosmos-tokenized dataset can be found 

On the horizon is our third challenge, Evaluation. This is our ultimate goal: can you predict how well a robot will perform before testing it in the real world? This challenge aims to assess the ability to evaluate and rank different robot policies using a world model, without the need for physical deployment.

The official details for the evaluation challenge have not been released yet—stay tuned for the announcement.

 - starter code, evals, baseline implementations


1X World Model: Sampling Challenge Update

 is a computer program that can imagine how the world evolves in response to an agent’s behavior. Building on advancements in 

, we have trained a world model that serves as a virtual simulator for our robots.

From the same starting image sequence, our world model can imagine multiple futures from different robot action proposals.

It can also predict non-trivial object interactions like rigid bodies, effects of dropping objects, partial observability, deformable objects (curtains, laundry), and articulated objects (doors, drawers, curtains, chairs).

In this post we’ll share why world models for robots are important, the capabilities and limitations of our current models, and a new dataset and public competition to encourage more research in this direction.

World models solve a very practical and yet often overlooked challenge when building general-purpose robots: evaluation. If you train a robot to perform 1000 unique tasks, it is very hard to know whether a new model has made the robot better at all 1000 tasks, compared to a prior model. Even the same model weights can experience a rapid degradation in performance in a matter of days due to subtle changes in the environment background or ambient lighting.

If the environment keeps changing over time, then old experiments performed in that environment are no longer reproducible because the old environment no longer exists! This problem gets worse if you are evaluating multi-task systems in a constantly-changing setting like the home or the office. This makes careful robotic science in the real world frustratingly hard.

Careful measurement of capabilities allows one to predict how capabilities will scale when one increases data, compute, and model size – these “

” defend the enormous investment that goes into general-purpose AI systems like ChatGPT. If robotics is to have its “ChatGPT moment”, we must first establish its “Scaling Laws”.

Physics-based simulation (Bullet, Mujoco, Isaac Sim, Drake) are a reasonable way to quickly test robot policies. They are resettable and reproducible, allowing researchers to carefully compare different control algorithms. However, these simulators are mostly designed for rigid body dynamics and require a lot of manual asset authoring. How to simulate robot hands opening a cardboard box of coffee filters, cutting fruit with a knife, unscrewing a frozen jar of preserves, or interacting with other intelligent agents like humans? Everyday objects and animals encountered in home environments are notoriously difficult to simulate, so simulation environments used in robotics tend to be visually sterile and lack the diversity of the real world use case. Small-scale evaluation on a limited number of tasks in real or sim is not predictive of large-scale evaluation in the real world.

We’re taking a radically new approach to evaluation of general-purpose robots: learning a simulator directly from raw sensor data and using it to evaluate our policies across millions of scenarios. By learning a simulator directly from real data, you can absorb the full complexity of the real world without manual asset creation.

Over the last year, we’ve gathered thousands of hours of data on EVE humanoids doing diverse mobile manipulation tasks in homes and offices and interacting with people. We combined the video and action data to train a world model that can anticipate future video from observations and actions.

Our world model is capable of generating diverse outcomes based on different action commands. Below we show various generations conditioning the world model on four different trajectories, each of which start from the same initial frames. As before, the examples shown are not included during training.

The main value of the world model comes from simulating object interactions. In the following generations, we provide the model the same initial frames and three different sets of actions to grasp boxes. In each scenario, the box(es) grasped are lifted and moved in accordance with the motion of the gripper, while the other boxes remain undisturbed.

We can also generate long-horizon videos. The example below simulates a complete t-shirt folding demonstration. T-shirts and deformable objects tend to be difficult to implement in rigid body simulators.

Our model can fail to maintain the shape and color of objects during interaction, and at times, objects may completely disappear. Additionally, when objects are occluded or displayed at unfavorable angles, their appearance can become distorted throughout the generation.

The generation on the left demonstrates that our model has an emergent understanding of physical properties, as evidenced by the spoon falling to the table when released by the gripper. However, there are many instances where generations fail to adhere to physical laws, such as on the right where the plate remains suspended in the air.

We placed EVE in front of a mirror to see if generations would result in mirrored actions, but we did not see successful recognition or “self-understanding"

As shown by the examples above, there is still much work to be done. World models have the potential to solve general purpose simulation and evaluation, enabling robots that are safe, reliable, and intelligent in a wide variety of scenarios. As such, we see this effort as a grand challenge in robotics that the community can work on solving together. To help accelerate progress towards solving world models for robotics, we are 

releasing over 100 hours of vector-quantized video

 (Apache 2.0), pretrained baseline models, and the 

, a three-stage challenge with cash prizes.

Compression Challenge | Prize: $10,000 USD

The first challenge, compression, is about how well one can minimize training loss on an extremely diverse robot dataset. The lower the loss, the better the model understands the training data. Even though there are many different ways to implement a world model, optimizing loss well is a general objective that underpins nearly all large-scale deep learning tasks. A $10k prize is awarded to the first submission that achieves a loss of 8.0 on our private test set. The Github repo provides code and pretrained 

The second challenge, sampling, is about how well and how quickly a model can generate videos of the future. Details of the Sampling Challenge will be announced soon, based on lessons learned from running the Stage 1 Challenge.

The third challenge, evaluation, is our holy grail: can you predict how well a robot performs before you test it in the real world? Details of the Evaluation Challenge will be announced after we’ve learned lessons from Stage 1 and Stage 2 Challenges.

If you’re excited about these directions, we have 

 on the 1X AI team. Internally, we have a large dataset of high resolution robot data across even more diverse scenarios. Our ambitions for world models go beyond just solving the general evaluation problem; once you can step an agent in this world model and perform evaluation, you can follow on with 

1X World Model

We have previously developed an autonomous model that can 

 goal-conditioned neural network. However, when multi-task models are small (<100M parameters), adding data to fix one task’s behavior often adversely affects behaviors on other tasks. Increasing the model parameter count can mitigate this forgetting problem, but also take longer to train, which slows down our ability to find out what demonstrations we should gather to improve robot behavior.

How do we iterate quickly on the data while building a generalist robot that can do many tasks with a single neural network? We want to decouple our ability to quickly improve task performance from our ability to merge multiple capabilities into a single neural network. To accomplish this, we’ve built a voice-controlled natural language interface to chain short-horizon capabilities across multiple small models into longer ones. With humans directing the skill chaining, this allows us to accomplish the long-horizon behaviors shown in this video:

Although humans can do long horizon chores trivially, chaining multiple autonomous robot skills in a sequence is hard because the second skill has to generalize to all the slightly random starting positions that the robot finds itself in when the first skill finishes. This compounds with every successive skill - the third skill has to handle the variation in outcomes of the second skill, and so forth.

From the user perspective, the robot is capable of doing many natural language tasks and the actual number of models controlling the robot is abstracted away. This allows us to merge the single-task models into goal-conditioned models over time. Single-task models also provide a good baseline to do 

 evaluations: comparing how a new model’s predictions differ from an existing baseline at test-time. Once the goal-conditioned model matches single-task model predictions well, we can switch over to a more powerful, unified model with no change to the user workflow.

Directing robots with this high-level language interface offers a new user experience for data collection. Instead of using VR to control a single robot, an operator can direct multiple robots with high level language and let the low-level policies execute low-level actions to realize those high-level goals. Because high-level actions are sent infrequently, operators can even control robots remotely, as shown below:

Note that the above video is not completely autonomous; humans are dictating when robots should switch tasks. Naturally, the next step after building a dataset of vision-to-natural language command pairs is to automate the prediction of high level actions using vision-language models like 

AI Update: Voice Commands & Chaining Tasks

Opt in to receive updates. Unsubscribe anytime.

Experience NEO

This move comes as we prepare for large-scale deployment of NEO, your friendly home humanoid. With the aim of having hundreds of NEO’s arrive in homes across the United States in 2025, followed by rapid expansion– Mustally will be instrumental in enabling our growth along the way.

“I am excited to welcome Mustally to the Executive Team at 1X,” said Bernt Børnich, CEO of 1X Technologies. “He will play a key role in helping us accelerate our growth plans from our global HQ in Palo Alto, California.”

Mustally brings a strong track record of leadership across Fortune 500 companies and high-growth startups. He has built finance organizations from the ground up, raised over $50 billion in capital through public and private markets, and led transformative commercial partnerships and M&A initiatives. Prior to joining 1X, Mustally served as Managing Director, Global Treasurer, and Head of Financial Services at Lucid Group, and held senior finance roles at Herc Holdings, Hyundai Capital, and National Grid, following a career in investment banking and management consulting.

“I’m thrilled to join the 1X team, which is well positioned to lead the physical AI and robotics sector,” said Mustally Hussain. “My focus will be on driving financial discipline and strategic growth as 1X scales its manufacturing and commercialization footprint in the U.S., followed by global expansion.”

Welcoming Mustally Hussain as CFO

NEO Gamma is the next generation of home humanoids designed and engineered by 1X Technologies. The Gamma series includes improvements across NEO’s hardware and AI, featuring a new design that is deeply considerate of life at home.


NEO Gamma's new design features Emotive Ear Rings to improve the communication and has a minimalist design aesthetic that fits into the home.




NEO can now walk with a natural human gait and arm swings, squat down to pick things up from the ground and sit in chairs.




NEO Gamma comes outfitted with soft covers to reduce impact on the surrounding environment and increase overall safety.




1X trained a visual manipulation model capable of picking up a large variety of objects in different scenarios, including environments not seen during training.




NEO Gamma’s Companion feature set integrates a new in-house language model that enables natural conversation and body language.

NEO Gamma’s design opens the door to start internal home testing—a first step in creating fully autonomous humanoids.

“There is a not-so-distant future where we all have our own robot helper at home, like Rosey the Robot or Baymax. But for humanoid robots to truly integrate into everyday life, they must be developed alongside humans, not in isolation.

The home provides real-world context and the diversity of data needed for humanoids to grow in intelligence and autonomy. It also teaches them the nuances of human life—how to open the door for the elderly, move carefully around pets, or adapt to the unpredictability of the surrounding world. Robots confined to industrial space or lab development miss out on this critical understanding.

With NEO Gamma, every engineering and design decision was made with one goal in mind: getting NEO into customers’ homes as quickly as possible… We’re close. We can’t wait to share more soon.”

NEO Gamma introduces a number of AI advancements, enhancing both autonomy and safe teleoperation in the home.




NEO now walks with a natural human gait and arm swings, squat down to pick things up from the ground and sit in chairs, all while maintaining balance. These dynamic control skills – running at 100Hz – are learned using Reinforcement Learning from human motion capture data. This range of motion allows NEO Gamma to experience household scenarios and tasks that would be otherwise inaccessible to other robot form factors.




1X trained a visual manipulation model capable of picking up a large variety of objects in different scenarios, including environments not seen during training. NEO Gamma leverages neural networks trained to predict teleoperated actions directly from raw sensor data.




NEO Gamma’s Companion feature set integrates a conversational voice interface, allowing users to converse naturally with a 1X developed language model (LLM). These developments bring NEO closer to human-friendly user interaction, bridging the gap between humanoid robotics and daily life.

NEO Gamma is designed to be approachable, with all new features created to integrate seamlessly into people’s everyday lives. NEO’s new look aims to complement living spaces rather than disrupt them.




NEO Gamma’s Ear Rings provide real-time visual feedback for a more intuitive and interactive experience.




3D printed from durable and soft nylon using a Japanese Shimaseki machine, the suit utilizes a unique whole-garment seamless knitting process that allows the fabric to conform to NEO without impeding performance.




With safety as the top priority, NEO Gamma features 1X’s Tendon Drive for joint actuation, encased in soft covers to enhance passive safety.

NEO Gamma features significant hardware upgrades, making it quieter, more reliable, and well suited to deliver a consumer product experience.




NEO Gamma boasts a 10x increase in hardware reliability and a 10 dB decrease in noise, bringing NEO’s operating levels down to the level of a refrigerator.




NEO Gamma features four microphones (front, back, left, and right) with beamforming and echo cancellation, ensuring crystal-clear audio capture and A three-speaker system—one in the chest for AI voice interactions and two outward-facing speakers in the pelvis for bass, 360-degree sound effects, and music.

Media Inquiries: press@1x.tech

Introducing NEO Gamma

 1X announces the acquisition of Kind Humanoid, uniting two robotics teams with aligned visions to advance humanoid technology.

“It’s rare to find someone who is not just a powerhouse engineer but also completely aligned philosophically and strategically on how humanoids as products should take shape,” said Bernt Bornich, CEO of 1X. “Having Christoph join the 1X team here in the Bay will accelerate our path to a world full of humanoid robots."

Kind Humanoid’s unique expertise and culture complement 1X’s mission to create an abundance of labor through safe and intelligent humanoids. The relationship between the two newly joined companies is built on the foundational belief that humanoids need to be developed alongside humans-- living and learning among us.

“Joining 1X feels like the perfect next chapter for Kind Humanoid,” said Christoph Kohstall, CEO of Kind Humanoid. “From starting out in a small garage to now becoming part of a team that shares our belief in humanoids living and learning among us, this acquisition brings our vision closer to reality. Together, we can create robots that truly connect with people and make a difference where it matters most.”

Kind Humanoid is a Palo Alto-based robotics startup founded by Christoph Kohstall, a former scientist at Stanford and member of Google's robotics team. The company developed Mona, a bipedal humanoid robot designed for home use and to bring additional labor to fields such as healthcare. Operating from a small garage, the team brought together a deeply bio-inspired robot body with advanced Al and large language models developing a robot that can interact with and assist everyday people.

1X is a leader in humanoid robotics, developing general-purpose robots designed to live and learn among humans in the home—an early step toward advancing physical intelligence. With a focus on safety, 1X is at the forefront of the next wave of robotics, offering products that are accessible and practical for everyday use. 1X’s mission is to create an abundance of labor through safe, intelligent humanoids that work alongside people

1X Acquires Kind Humanoid

Our goal with this video was to create a fun and exciting glimpse into a not-so-distant future where humanoids cook delicious meals for us.

After connecting with Nick and his team, we brainstormed ideas for an engaging narrative. When Nick suggested a robot cook-off with NEO Beta as the “final boss,” we quickly set to work refining our cooking skills.

While cooking a steak dinner from start to finish was an impressive demonstration of NEO Beta's ability to navigate complex tasks, it is important to note that this showcase was done using teleoperation to ensure the successful execution of all tasks involved. That said, the only thing standing between NEO and a fully autonomous, medium-rare steak is the input data.

In the video, NEO Beta and Nick met up in Sunnyvale, California, near the 1X headquarters, for a home cooking showdown to see who could make the perfect medium-rare steak. The audience would ultimately decide the winner.

Nick generously brought NEO Beta a custom “Chef NEO” coat and stocked the fridges with more Wagyu beef than any humanoid could ever need.

: We (wrongly) assumed NEO Beta might mess up at least one steak since this was an entirely new skill. To be safe, we had multiple backup steaks on hand. Proving us wrong, NEO Beta nailed it on the first try, and the 1X team and Nick got to enjoy the extra steaks afterward.

The shoot was full of laughs, imperfections like knocking over the olive oil, and memorable moments like when NEO Beta pulled off butter basting—reminding everyone on set they were witnessing a landmark moment in the history of home robotics.

: Nick admitted that NEO Beta was a better chef than Gordon Ramsay and Uncle Roger. He asked us to keep it a secret, but we couldn’t resist sharing.

While videos like this are extremely fun and important for showing the masses that humanoid robots are here and the fun is just beginning– we find it equally important to be transparent about how we capture these shots to maintain trust in the robotics community and set the right expectations for our product.

All movements were powered by 1X’s VR Teleoperation App running on Meta Quest.

NEO Beta's lines were filmed using voice pass-through with a filter. Although NEO is equipped with real-time conversation capabilities via our GPT-4o voice integration, we opted for controlled dialogue to align with Nick’s vision for the video.

NEO Beta’s voice was designed in collaboration with Nick’s team to find a tone that was playful and appealing to both his audience and our vision of NEO as a friendly, home-assisting humanoid.

: Although NEO Beta successfully cooked a meal alongside Nick, cooking won’t be an immediate feature available to the first NEO users. We want to ensure NEO gains experience with safer tasks before handling sharp or hot objects.

: While NEO Beta completed all tasks from seasoning the steak to flipping and removing it from the pan end to end, NEO Beta did require assistance to turn on the burner. Nick and his team felt that including this scene was important for their audience, so we agreed to create this article for transparency.

We’re incredibly excited that NEO Beta was able to cook a full steak dinner with minimal assistance. Sooner than you think, NEO will be cooking similar meals—and more—in your kitchen.

Huge thanks to Nick, Tim and Zach for putting this all together.

We hope you enjoyed the video. Share it with friends and family and let us know what you all think on 

Redwood AI: Mobility

Redwood AI

1X & NVIDIA Research Collaboration

1X World Model: Sampling Challenge Update

1X World Model

ExperienceNEO