CtrlK
| 1X NEO Beta |
|---|
| General information |
| Manufacturer |
| Country of origin |
| Year introduced |
| Status |
| Height |
| Weight |
| Degrees of freedom |
| Actuation |
| Max walking speed |
| Max running speed |
| Payload capacity |
| Battery life |
| Noise level |
| Website |
1X NEO Beta is a bipedal humanoid robot developed by 1X Technologies (formerly Halodi Robotics), a Norwegian-American robotics company. Unveiled on August 30, 2024, in San Francisco and Moss, Norway, NEO Beta was the first physical, walking prototype of the 1X Neo platform. It served as an engineering validation unit built to prove that 1X's bio-inspired design philosophy, centered on lightweight construction, tendon-driven actuation, and a soft exterior, could translate from concept into a functional bipedal system capable of operating inside private homes.[1][2]
Standing 165 cm tall and weighing just 30 kg, NEO Beta was dramatically lighter than competing humanoids such as Tesla Optimus (73 kg) and Figure 02 (70 kg). The robot combined 55 degrees of freedom, hands with 22 degrees of freedom each, a soft cushioned body, and tendon-driven actuators to produce movements described as gentle and compliant rather than stiff or jerky. NEO Beta demonstrated walking at up to 4 km/h and sprinting at up to 12 km/h, with a payload capacity of 20 kg.[3]
As a prototype, NEO Beta was never sold commercially. Instead, 1X deployed a limited number of units in employees' homes and with select early testers during late 2024 to gather real-world feedback. The data and lessons collected from NEO Beta directly informed the development of NEO Gamma, which was unveiled in February 2025 as the consumer-ready successor with a tenfold improvement in hardware reliability.[4][5]
1X Technologies was founded in May 2014 in Moss, Norway, by roboticist Bernt Oivind Bornich along with co-founders Phuong Nguyen, Jorgen Sundell, and Pal Loken. The company was originally named Halodi Robotics and focused on developing safe actuators and full-body control systems for robots that could work alongside humans. Its earliest innovation was the Revo1 motor, a high-torque brushless direct current (BLDC) motor paired with low-gear-ratio cable drives. At the time of its development, the Revo1 was described as the highest torque-to-weight direct drive motor in the world.[6]
In 2018, the company released EVE, a wheeled humanoid robot standing 188 cm tall and weighing 87 kg. EVE was designed for logistics, security, and healthcare applications. Its most significant commercial deployment came through a partnership with ADT Commercial, which ordered between 140 and 250 EVE units for autonomous night security patrols in commercial buildings across the United States.[7]
In March 2023, Halodi Robotics rebranded as 1X Technologies and pivoted from enterprise robotics toward the consumer market. That same month, the company closed a $23.5 million Series A2 funding round led by the OpenAI Startup Fund, with participation from Tiger Global and Norwegian investors, at a post-money valuation of approximately $210 million. This round marked one of OpenAI's first major investments in physical robotics and established a strategic partnership giving 1X early access to OpenAI's AI research and models.[8][9]
In January 2024, 1X secured $100 million in Series B funding led by EQT Ventures, with participation from Samsung NEXT, the Nistad Group, and other investors. This brought total funding to over $125 million and gave the company the capital to scale manufacturing at its Norway facility and hire aggressively for the NEO program.[10]
The NEO platform was first shown publicly at NVIDIA's GTC AI conference in March 2024, during Jensen Huang's keynote address. At GTC, NEO appeared alongside humanoid robots from other major robotics companies in a demonstration of the growing humanoid ecosystem. At this stage, NEO was still a concept; the robot shown at GTC did not demonstrate autonomous bipedal walking or manipulation in a live setting.[11]
Between March and August 2024, 1X's engineering team in Moss and Sunnyvale worked to transform the NEO concept into a functional walking prototype. The team needed to validate several core technologies simultaneously: the Tendon Drive actuation system (previously proven only in the wheeled EVE platform), bipedal locomotion control, the soft-body safety design, and the sensor and compute stack required for autonomous navigation in unstructured home environments.[2]
On August 30, 2024, 1X held simultaneous events in San Francisco Bay and Moss, Norway, to unveil NEO Beta as the first physical prototype of the NEO humanoid platform. CEO Bernt Bornich emphasized that safety was the primary design constraint, stating: "Safety is the cornerstone that allows us to confidently introduce NEO Beta into homes, where it will gather essential feedback and demonstrate its capabilities in real-world settings."[1]
The unveiling event showcased NEO Beta walking bipedally, grasping objects, and performing basic household tasks. The company released video footage of the robot in a domestic setting, although 1X did not specify at the time which demonstrated behaviors were fully autonomous, teleoperated, or pre-programmed routines.[3]
Alongside the hardware reveal, 1X announced a breakthrough in its artificial intelligence approach: the 1X World Model (1XWM). This was a 14-billion-parameter generative video model that could convert text-conditioned video generation into robot actions. Rather than training a traditional vision-language-action model that predicts action trajectories from static images, the world model leverages physical dynamics learned from internet-scale video data to generalize to novel objects and tasks. The model was trained on 900 hours of egocentric human video and fine-tuned on 70 hours of robot-specific data to adapt to NEO's visual appearance and kinematics.[12]
NEO Beta was designed with one overriding principle: a humanoid robot intended for homes must be inherently safe around people, pets, and furniture without relying solely on software safeguards. This led to a design that prioritized low mass, soft materials, compliant actuation, and quiet operation. The approach stood in contrast to most competing humanoids, which were designed for industrial or warehouse environments and typically weighed 60 to 90 kg with rigid metal exteriors.[2]
At 30 kg, NEO Beta carried significantly less kinetic energy during movement than heavier robots, reducing the severity of any accidental collision. The robot's exterior featured cushioned inserts positioned where a human body would have muscles, eliminating pinch points at joints and creating a forgiving contact surface. The overall form factor was designed to be human-proportioned so that the robot could navigate standard doorways, walk through furnished rooms, and interact at a natural eye level with adults.[3]
The most distinctive engineering feature of NEO Beta was its Tendon Drive actuation system, based on 1X's patented technology originally developed for the EVE platform. Instead of conventional gearboxes or harmonic drives, the system uses high-torque-density brushless DC motors connected to cable-driven differential transmissions. This configuration mimics the tendon-and-muscle arrangement found in the human musculoskeletal system.[6]
The Tendon Drive approach provides several advantages over rigid gear-driven actuators:
| Characteristic | Description |
|---|---|
| Backdrivability | At 95% backdrivability, if a person pushes against the robot, its joints yield smoothly rather than resisting with rigid force |
| Low noise | The tendon system operates with minimal mechanical noise, contributing to quiet operation |
| Compliance | Movements are smooth and organic rather than stiff and jerky, which is both safer and less psychologically jarring |
| Power density | The Revo1 motors deliver high torque relative to their weight, enabling a lightweight robot to carry meaningful payloads |
The team at 1X rejected the conventional approach of using high-ratio harmonic gears, which introduce friction, reflected inertia, and limited controllability. By pairing high-torque motors with low-ratio cable drives, the Tendon Drive system achieved a combination of strength and gentleness that was central to the safety-first design philosophy.[6][13]
NEO Beta's hands featured 22 degrees of freedom per hand, providing human-level dexterity for grasping and manipulating household objects. The five-fingered design was engineered to handle a wide range of object shapes, sizes, and weights, from delicate items like eggs to heavier objects like pots and pans. The hands were rated IP68, meaning they were fully submersible in water, an important feature for a robot intended to assist with kitchen tasks and cleaning.[3][14]
NEO Beta was equipped with a comprehensive sensor array designed to provide the perception needed for autonomous operation in cluttered, dynamic home environments:
| Sensor | Details |
|---|---|
| Vision | Dual 8 MP fisheye RGB-D cameras with 180-degree field of view |
| Depth sensing | LiDAR for spatial mapping and obstacle detection |
| Inertial measurement | IMUs with linkwise differential sensing for body state estimation |
| Force/torque | Force-torque sensors at key joints for compliant interaction |
| Position tracking | Position encoders at each joint |
| Audio | Four microphones with beamforming capability |
| Proximity | Proximity sensors for close-range obstacle avoidance |
The dual fisheye cameras provided near-hemispherical visual coverage, allowing NEO Beta to perceive its surroundings without needing to turn its head frequently. The LiDAR sensor, combined with visual odometry from the cameras, supported SLAM (Simultaneous Localization and Mapping) for autonomous navigation through rooms and hallways.[3][14]
NEO Beta ran on a Linux-based operating system with an ROS 2 framework. The onboard AI stack included 1X's Redwood vision-language-action model, a 160-million-parameter neural network capable of interpreting visual scenes and executing manipulation tasks end-to-end. For navigation, the robot combined LiDAR-based SLAM with visual odometry to build and maintain real-time maps of its environment.[12][14]
The robot supported connectivity through Wi-Fi, Bluetooth, and cellular (5G) radios, with integration into the 1X Cloud platform for telemetry, analytics, and over-the-air software updates. A companion mobile app, the 1X Home App, allowed users to schedule tasks and monitor the robot remotely.[14]
| Category | Specification | Value |
|---|---|---|
| Physical | Height | 165 cm (5 ft 5 in) |
| Width | 45 cm | |
| Depth | 30 cm | |
| Weight | 30 kg (66 lb) | |
| Mobility | Total degrees of freedom | 55 |
| Degrees of freedom per hand | 22 | |
| Max walking speed | 1.1 m/s (4.0 km/h) | |
| Max running speed | 3.3 m/s (12.0 km/h) | |
| Strength | Carrying capacity (per arm) | 20 kg |
| Maximum deadlift | 70 kg | |
| Power | Battery type | Lithium-ion |
| Estimated battery capacity | ~1.2 kWh | |
| Runtime | 2 to 4 hours | |
| Charging time | ~2 hours to full (docking station) | |
| Average power draw | ~150 W | |
| Sensors | Cameras | Dual 8 MP fisheye RGB-D (180-degree FOV) |
| LiDAR | Yes | |
| IMU | Linkwise differential | |
| Force-torque sensors | Yes (key joints) | |
| Microphones | 4 (beamforming capable) | |
| Proximity sensors | Yes | |
| Compute | Operating system | Linux-based with ROS 2 |
| AI model | Redwood VLA (160M parameters) | |
| Navigation | SLAM (LiDAR + visual odometry) | |
| Safety | Noise level | ~32 dB |
| Body IP rating | IP44 (splash-proof) | |
| Hand IP rating | IP68 (submersible) | |
| Emergency stop | Yes | |
| Collision detection | Yes |
Before NEO Beta existed as a walking prototype, the NEO concept appeared at NVIDIA's GTC conference in March 2024. During Jensen Huang's keynote, 1X's robot was featured alongside humanoids from other companies in a demonstration highlighting the emerging humanoid robotics ecosystem. This appearance generated significant public attention for the NEO program, even though the robot shown at GTC was a pre-Beta concept unit rather than a fully functional prototype.[11]
The formal unveiling on August 30, 2024, included demonstrations of NEO Beta walking, navigating a domestic interior, and grasping objects. Video footage released by 1X showed the robot moving through a kitchen and living room setting, picking up items from countertops and tables, and responding to voice commands. The demonstration highlighted the robot's quiet operation and the smooth, natural quality of its tendon-driven movements.[1][2]
In November 2024, NEO Beta appeared in a widely viewed video collaboration with YouTube chef Nick DiGiovanni, filmed in Sunnyvale, California, near 1X's then-headquarters. In the video, NEO Beta and DiGiovanni competed in a home cooking showdown to prepare the perfect medium-rare steak.[15]
All of NEO Beta's movements during the cooking demonstration were controlled through 1X's VR Teleoperation App running on a Meta Quest headset. A human operator guided the robot's actions in real time, rather than the robot acting autonomously. NEO Beta successfully completed the full cooking process: seasoning the steak, placing it in a hot pan, flipping it, performing butter basting, and removing the finished steak. The robot required human assistance only for turning on the gas burner. NEO Beta's spoken lines were generated using voice pass-through with a filter applied to the operator's voice.[15]
The video accumulated over 40 million views across social media platforms, making it one of the most widely seen humanoid robot demonstrations of 2024. However, 1X noted that cooking would not be an immediately available capability for consumer NEO units, as the company intended to validate the robot on safer tasks before allowing it to handle sharp objects or hot surfaces.[15][16]
Following the public unveiling, 1X deployed a limited number of NEO Beta units in selected homes for research and development during late 2024. These deployments were not commercial sales but internal pilot programs designed to collect real-world data on the robot's performance in unstructured domestic environments. The feedback from these pilots covered hardware durability, locomotion stability on various floor surfaces, manipulation reliability with everyday household objects, and the psychological comfort level of human occupants sharing living space with a bipedal robot.[4][5]
The pilot program revealed areas where NEO Beta needed improvement, including hardware reliability, noise levels during extended operation, and the naturalness of the robot's walking gait. These findings became the primary engineering targets for the next iteration of the platform.[5]
Alongside NEO Beta, 1X introduced the 1X World Model, a 14-billion-parameter generative video model that represented a fundamentally different approach to robot learning. Traditional robot AI systems use vision-language-action (VLA) models that predict action trajectories directly from static image and language inputs. The 1XWM instead derives robot actions from text-conditioned video generation, leveraging the rich physical dynamics embedded in internet-scale video data.[12]
The training strategy for the world model proceeded in multiple stages. First, the model was pre-trained on large-scale internet video to learn general physical dynamics. Second, it was fine-tuned on 900 hours of egocentric human video captured from a first-person perspective. Third, it was further refined on 70 hours of robot-specific data to adapt the model to NEO's particular visual appearance and kinematics. During inference, the model generates a short video clip in approximately 11 seconds and then extracts robot actions from the generated video in about 1 second.[12]
This approach allowed the world model to generalize to novel objects, movements, and tasks that the robot had never encountered during training, without requiring large-scale teleoperated demonstration data for each new task.
NEO Beta also ran 1X's Redwood AI system, a vision-language-action model with 160 million parameters. Redwood served as the robot's generalist AI, enabling it to interpret visual scenes, understand natural language commands, and execute manipulation tasks end-to-end. The system was trained on raw sensor data from NEO's onboard cameras and was capable of picking up a variety of objects in different scenarios, including environments not seen during training.[12][14]
Redwood was designed to improve over time as more robots were deployed and generated additional training data, forming a data flywheel where each deployed unit contributed to improving the AI capabilities of all units in the fleet.
NEO Beta was explicitly designed as an engineering validation platform rather than a consumer product. Its purpose was to prove that the core NEO design concepts worked in practice and to identify the areas requiring improvement before the platform could be manufactured and sold at scale.[4]
On February 21, 2025, 1X introduced NEO Gamma, the consumer-ready evolution of the platform. NEO Gamma incorporated extensive improvements informed by the Beta pilot program, addressing nearly every subsystem of the robot. The following table compares the two versions across key dimensions:
| Feature | NEO Beta | NEO Gamma |
|---|---|---|
| Role | Engineering prototype | Consumer-ready design |
| Height | 165 cm | 165 cm |
| Weight | 30 kg | 30 kg |
| Total degrees of freedom | 55 | 75 |
| Degrees of freedom per hand | 22 | 22 |
| Max walking speed | 4 km/h (1.1 m/s) | 5 km/h (1.4 m/s) |
| Max running speed | 12 km/h (3.3 m/s) | 22.3 km/h (6.2 m/s) |
| Carrying capacity | 20 kg | 25 kg |
| Lift capacity | 70 kg | 70 kg |
| Battery capacity | ~1.2 kWh | 842 Wh (optimized) |
| Runtime | 2 to 4 hours | ~4 hours (typical) |
| Hardware reliability | Baseline | 10x improvement over Beta |
| Operating noise | ~32 dB | ~22 dB (10 dB reduction) |
| Exterior covering | Exposed frame with cushioned inserts | Seamless knit nylon bodysuit over 3D lattice polymer |
| Gait quality | Basic bipedal walking | Natural human gait with arm swings |
| Movement capabilities | Walking, basic grasping | Squatting, sitting in chairs, ground pickup |
| Audio system | 4 microphones, basic speakers | 4-mic beamforming array with echo cancellation, 3-speaker system |
| Locomotion control | Standard | Reinforcement learning from motion capture at 100 Hz |
| Compute | Linux/ROS 2 stack | 1X NEO Cortex (NVIDIA Jetson Thor, Blackwell architecture) |
| Visual feedback | None | Emotive Ear Rings (LED status indicators) |
| Exterior colors | N/A (prototype) | Tan, gray, dark brown |
The most dramatic improvements from Beta to Gamma were in running speed (nearly doubled), hardware reliability (10x), noise reduction (10 dB quieter), the addition of 20 more degrees of freedom across the body, and the upgrade to NVIDIA Jetson Thor compute hardware. The Gamma version also introduced a complete aesthetic redesign led by former BMW designer Per Selvaag, with the goal of making the robot psychologically acceptable for daily cohabitation with humans.[4][5][17]
In January 2025, between the Beta and Gamma milestones, 1X acquired Kind Humanoid, a Palo Alto-based robotics startup founded by Christoph Kohstall, a former scientist at Stanford University and member of Google's robotics team. Kind Humanoid had been developing Mona, a bipedal humanoid robot designed for home use and healthcare applications. The acquisition brought additional design and engineering talent into 1X's team and accelerated the development of NEO Gamma and subsequent production models.[18]
NEO Beta occupies a pivotal position in the history of the 1X Neo platform. It transformed the NEO concept from a conference-stage prototype into a validated walking, grasping, household-capable humanoid. While NEO Beta was never intended for commercial sale, it accomplished several objectives that were critical for the platform's progression:
The production version of NEO, based on the Gamma design, opened for consumer pre-orders on October 28, 2025, at an Early Access price of $20,000, with first U.S. deliveries scheduled for 2026.[19]