Robots were blind to common-sense semantic context; they were simply repeaters of fixed geometric math.

The rise of Large Language Models (LLMs) and Vision-Language Models (VLMs) like GPT-4o, Claude, and Gemini demonstrated that neural networks trained on internet-scale text and images could master semantic reasoning, visual scene decomposition, and abstract logic. Yet, these models remained “disembodied.” A large vision-language model could look at a photo of a kitchen counter, correctly identify an apple, and write a detailed poetic essay on how to slice it—yet it possessed zero physical agency to output the joint torques, velocity vectors, and end-effector gripper states needed to move an arm across space.

To bridge this physical chasm, roboticists created Vision-Language-Action (VLA) models.

Pioneered by Google DeepMind with RT-2 (Robotics Transformer 2) and rapidly expanded by open-source frameworks like OpenVLA and industrial foundation engines like Figure AI’s Helix, VLAs represent the foundational operating system of modern physical AI.

By unifying visual perception tokens, natural language prompt tokens, and physical actuator action vectors into a single end-to-end neural network, VLAs allow humanoids to convert spoken human intent (“Hand me the tool that can cut this zip-tie”) directly into fluid, closed-loop physical movements without human teleoperation or pre-scripted trajectories.

Key Architectural Takeaways

• The Unified Token Paradigm: VLAs treat physical motor commands not as isolated numerical outputs, but as discrete “action tokens” or continuous latent embeddings integrated directly into the language model’s vocabulary.

• Internet-Scale Semantic Transfer: By building on top of pre-trained VLM backbones, robots inherit common-sense physical reasoning (e.g., knowing that an energy drink helps someone who is tired, or that a rock can be used as an improvised hammer) without needing explicit robot demonstration data for every concept.

• Action Prediction Strategies: Architectures split between autoregressive discrete tokenizers (RT-2, OpenVLA, predicting 256 discretized bins per degree of freedom) and continuous generative decoders (Diffusion Policy / Flow Matching action heads).

• Frequency Hierarchies (System 1 vs. System 2): Solves the compute latency bottleneck by separating high-level semantic reasoning (3 Hz to 10 Hz) from high-frequency whole-body kinematic balance and motor impedance control (200 Hz to 1,000 Hz).

• Multi-Embodiment Generalization: Datasets like the Open X-Embodiment (OXE) pool millions of trajectories across disparate robot types, enabling a single VLA backbone to generalize across single-arm grippers, quadrupeds, and full-scale 50-DoF bipedal humanoids.

Quick Specs: Foundational VLA Models Comparison

Internal Anatomy: How Pixels and Text Become Joint Torques

To understand a VLA, one must trace the flow of information through the network layers. Unlike traditional robotics architectures that maintain separate software modules for object recognition, path planning, and trajectory generation, a VLA executes these functions inside a unified neural transformer:

Stage 1: Multi-View Visual Encoding (The Vision Backbone)

• Head-mounted stereo cameras and wrist-mounted micro-cameras capture live video frames (e.g., $224 \times 224$ or $384 \times 384$ pixels).

• A pre-trained Vision Transformer (ViT) (such as SigLIP, CLIP, or Dinov2) breaks the 2D image into a grid of non-overlapping image patches (e.g., $14 \times 14$ pixels each).

• Linear projection layers convert these spatial patches into dense visual embedding tokens, encoding spatial relationships, object textures, lighting gradients, and depth cues.

↓ (Multimodal Sequence Fusion)

Stage 2: Language Context & Cross-Attention (The Transformer Core)

• The natural language task prompt (“Gently pick up the ripe peach from the crate and place it in the soft bowl”) is tokenized into standard text tokens via an LLM tokenizer (such as LLaMA or PaLM).

• The visual tokens and text tokens are concatenated into a single sequential input stream.

• Self-attention and cross-attention layers across the multi-billion-parameter backbone compute attention weights between words and pixels. The token representing “peach” attends directly to the visual tokens representing the fuzzy orange object, while “soft bowl” attends to the target container.

↓ (Action Generation & De-Tokenization)

Stage 3: Action Head Execution (The Physical Actuation Boundary)

• The model reaches its output projection layers. Instead of generating a next-token word in a sentence, the VLA outputs an action chunk.

• The action is expressed either as discrete tokens (mapped to $x, y, z$ positional deltas, roll-pitch-yaw rotations, and gripper clamp width) or as continuous latent vectors decoded by a downstream policy.

• These output coordinates pass to the low-level controller, which executes the joint movements.

Action Representation: Discrete Tokens vs. Continuous Diffusion Heads

The primary architectural debate within VLA engineering centers on how the network outputs physical actions. Physical motion is continuous and smooth, whereas language models are built to predict discrete, categorical text tokens.

Methodology 1: Discrete Action Tokenization (RT-2 & OpenVLA Approach)

• The Technique: The continuous range of each actuator or end-effector degree of freedom (e.g., an arm moving from -1.0 m to +1.0 m along the X-axis) is split into a discrete number of uniform buckets—typically 256 bins.

• Vocabulary Expansion: 256 reserved token IDs are appended to the language model’s text vocabulary dictionary.

• The Process: When the model outputs an action, it simply generates a string of text tokens representing numbers:

• Advantage: Simplicity. It requires zero architectural modifications to standard transformer decoders, allowing the model to be trained using cross-entropy loss, identically to predicting the next word in an essay.

• Disadvantage: Discretization error. Quantizing continuous physical motion into 256 steps creates stepping artifacts. Furthermore, predicting actions autoregressively token-by-token increases computational latency, capping control frequencies at a sluggish 3 Hz to 10 Hz.

↓ (Generative Paradigm Shift)

Methodology 2: Continuous Flow-Matching & Diffusion Policy Heads (Helix & Modern VLA Approach)

• The Technique: The transformer backbone does not predict raw action tokens. Instead, it outputs a rich conditioning latent vector that conditions a Diffusion Policy or Flow-Matching action head.

• The Process: Starting from pure Gaussian random noise, the action head runs a brief denoising process (over 4 to 10 denoising steps) to generate a smooth, continuous trajectory chunk across time:

• Multimodal Distributions: If an obstacle blocks an arm, a discrete token model can average two valid paths, causing the arm to plunge straight into the obstacle. Diffusion heads naturally model multi-modal probability distributions: they can decide to go 100% to the left OR 100% to the right with zero path averaging.

• Continuous Dynamics: Produces fluid, jitter-free joint movements running at high continuous frequencies, eliminating jerky stepping artifacts at the motor level.

The Frequency Problem: Decoupling System 1 and System 2

A massive challenge in deploying VLAs onto physical humanoids is computational latency. A 7-billion-parameter multimodal neural network takes 100 to 300 milliseconds to complete a forward pass on a high-end mobile GPU.

If a robot’s whole-body balance and obstacle-avoidance reflex loop ran at only 3 Hz, the robot would fall over the moment its foot stepped on an uneven pebble.

To solve this mismatch between cognitive reasoning and dynamic physics, modern physical AI implements a dual-rate hierarchical architecture (loosely modeled on Daniel Kahneman’s “System 1 and System 2” cognitive paradigm):

System 2: The High-Level Cognitive Brain (Slow / Asynchronous: 3 Hz to 10 Hz)

• Compute Engine: Multimodal VLA Foundation Model (e.g., OpenVLA, Helix VLA) running on edge GPUs (NVIDIA Jetson Thor / Blackwell).

• Sensory Input: Multi-camera RGB video, high-level natural language instructions, spatial audio cues.

• Task Scope: Semantic scene interpretation, object identification, sub-goal generation, and rough Cartesian trajectory chunking (“Reach toward the red lever, hold clamp open”).

↓ (Asynchronous Intermediate Action Trajectory Buffer)

System 1: The Low-Level Reflexive Spine (Fast / Deterministic: 200 Hz to 1,000 Hz)

• Compute Engine: Real-Time Embedded Microcontrollers (ARM Cortex-R / RTOS / FPGA / EtherCAT Master).

• Sensory Input: Joint optical encoders, 6-axis force/torque sensors, IMUs, tactile finger skins.

• Task Scope: Whole-Body Control (WBC), dynamic Zero-Moment Point (ZMP) balancing, Inverse Kinematics (IK), and active mechanical impedance modulation.

• Fail-Safe Reflexes: If a human worker brushes against the robot’s forearm while the VLA is mid-reach, System 1 detects the resistive current spike within 1.5 milliseconds, instantly softening joint stiffness or freezing movement without waiting for the slow System 2 model to process the visual frame.

Web-Scale Pre-Training to Physical Manipulation: Emergent Semantic Reasoning

Why not simply train a small convolutional neural network directly on robot demonstration data? Why use massive multi-billion-parameter vision-language models at all?

The breakthrough of VLAs lies in semantic transfer and emergent capabilities.

When a policy is trained strictly on robot teleoperation data, its knowledge is limited to the exact physical items it encountered during training. If a robot was trained to pick up yellow plastic cups, placing a translucent glass tumbler in front of it often causes the policy to fail completely.

Pre-Training Transfer Layer 1: Zero-Shot Object Generalization

• Because the VLA’s vision-language backbone was pre-trained on billions of image-text pairs from the web, it already knows what thousands of objects look like—even if it has never physically touched them on a robot.

• It understands visual concepts like “translucent,” “fragile,” “rusty,” or “folded” without needing robotic demonstrations for each attribute.

↓ (Cognitive Abstraction Expansion)

Pre-Training Transfer Layer 2: Abstract Reasoning and Chain-of-Thought

• In DeepMind’s RT-2 evaluations, the model was instructed: “Pick up the improvised hammer”.

• The robot was never trained on a demonstration labeled “improvised hammer.” Yet, leveraging its web-scale language knowledge, the VLA reasoned that a hammer requires mass and hardness, visually identified a rock resting on the table, and output the correct grasping trajectory to pick up the rock.

• Similarly, when told to “Clean up the spill,” the model bypassed nearby apples and plastic cups to grasp a porous sponge, understanding functional tool affordances purely through semantic transfer.

Operational Video Reference: VLA In-Action Demonstration

The real-time operational execution of generalist Vision-Language-Action models—unifying natural language understanding with multi-DoF upper-body humanoid coordination—is visible in active commercial platform milestones:

Figure Helix VLA Foundation Model Operational Showcase:

Watch the multimodal VLA execute collaborative manipulation tasks: Figure Helix: A Vision-Language-Action Model for Generalist Humanoid Control

• Key Observation Points:

  • Real-time semantic task decomposition from open-ended natural language prompts without pre-scripted code.

  • Coordinated continuous multi-DoF upper-body motion (arms, wrists, and individual fingers moving simultaneously).

  • Zero-shot grasping of novel household items and groceries never encountered during teleoperation training.

  • Synchronized multi-robot collaboration: two humanoids executing shared long-horizon assembly using a single neural model.

Commercial ROI: How VLAs Alter the Economics of Robot Programming

In classical industrial automation, the hardware represents only 25% to 35% of total project costs. The remaining 65% to 75% is consumed by systems integration: hiring teams of specialized robotics programmers to manually script trajectories, calibrate vision fixtures, design custom mechanical part feeders, and tune safety PLC interlocks.

If a manufacturing plant shifts from producing Sedans to SUVs, reprogramming the workcells takes months of downtime and millions of dollars in engineering billable hours.

Economic Vector 1: The Exponential Cost of Traditional Scripting

• Every new SKU or assembly variation requires bespoke trajectory coding and testing.

• Scaling to thousands of unstructured tasks in homes or warehouses is economically impossible through human programming alone.

↓ (Software Paradigm Disruption)

Economic Vector 2: The VLA Zero-Shot Scaling Curve

• New tasks are assigned simply by speaking natural language commands or providing a single visual goal prompt.

• A humanoid running an enterprise VLA can be moved from a packaging line to a kitting station without rewriting code. The foundation model inspects the novel workspace, identifies the bins, and initiates manipulation immediately based on broad physical priors.

• Deployment Velocity: Reduces onboarding time for new robotic manufacturing workflows from months down to hours, fundamentally disrupting the economics of factory and warehouse automation.

Engineering Verdict & Field Evaluation

Vision-Language-Action (VLA) Models: Pros & Operational Strengths

• Universal Semantic Generalization: Inherits web-scale common-sense knowledge, enabling robots to understand novel objects, spatial relationships, and open-ended language prompts without task-specific training.

• Unified End-to-End Learning: Replaces fragile multi-component software pipelines (object detection -> segmentation -> pose estimation -> motion planning) with a single, differentiable neural network.

• Intuitive Human Collaboration: Allows factory floor workers without computer science degrees to command, retask, and supervise robotic operations using spoken natural language.

Vision-Language-Action (VLA) Models: Limitations & Engineering Risks

• Inference Compute Overhead: Running 7B+ parameter models locally demands high-wattage edge GPUs (100 W to 300 W), consuming precious onboard battery reserves.

• The “Black Box” Explainability Problem: Unlike deterministic kinematics code, end-to-end neural networks cannot offer 100% formal safety proofs; debugging why a robot dropped a part or executed an erratic swing remains challenging.

• High-Precision Insertion Gaps: While VLAs excel at coarse grasping and object sorting, sub-millimeter industrial assembly (e.g., tight-tolerance bearing press-fitting) still requires localized force-torque feedback and classical compliance loops.

The Bot.to Benchmark Verdict:

Vision-Language-Action (VLA) models are the defining software architecture of the embodied AI era. They solve the foundational limitation of classical robotics: the inability to generalize across the messy, unstructured visual world.

However, VLAs will not replace classical robotics entirely. The winning commercial architecture is a hybrid mechatronic stack: massive multimodal VLA foundation models acting as the high-level semantic brain (System 2), feeding goal trajectories down into deterministic, gigahertz whole-body controllers and tactile impedance loops (System 1) that guarantee physical balance and millisecond-level human safety.

Frequently Asked Questions (FAQ)

Q: What does VLA stand for in robotics, and how is it different from an LLM?

A: VLA stands for Vision-Language-Action. While a Large Language Model (LLM) takes text in and outputs text, a VLA takes both camera images (Vision) and natural language text (Language) as inputs, and directly outputs low-level physical motor commands, joint angles, or end-effector trajectories (Action) to move a robot’s physical body.

Q: Can a VLA model run directly on a humanoid robot without an internet connection?

A: Yes, modern production-grade VLAs (such as Figure’s Helix or optimized 7B-parameter models like OpenVLA) are specifically optimized to run locally on embedded onboard AI hardware (such as NVIDIA Jetson Thor or dual automotive-grade GPUs). This eliminates cloud network latency, data privacy vulnerabilities, and the risk of the robot freezing if factory Wi-Fi drops out.

Q: What is the difference between discrete action tokens and continuous diffusion action heads?

A: Discrete action token models (like RT-2) break physical motion into discrete buckets (e.g., 256 numerical steps per joint) and predict them one by one like words in a sentence. Continuous diffusion heads use generative denoising to output complete, fluid, multi-joint trajectory curves all at once, eliminating robotic motor stutter and allowing smooth whole-body humanoid coordination.

Q: How do VLA models know what to do with objects they have never seen before?

A: VLAs inherit their visual and semantic understanding from vision-language backbones pre-trained on billions of photos and descriptions from the internet. Because the model has already “seen” millions of variations of cups, tools, fruits, and furniture online, it understands what they are and how they function physically, transferring that common sense to physical manipulation without needing bespoke robot training demos.

Explore related platforms and technical profiles in the Bot.to Humanoid Directory or read our direct hardware breakdown: Figure 02 vs. Tesla Optimus: Actuation, AI Stack, and Factory Deployment Timelines.

The domestic living room destroys every single one of those industrial assumptions.

In a private residence, robots do not operate behind safety cages. They navigate across hardwood floors and deep shag rugs, brush past toddlers, interact with pets, and handle fragile glassware. If a traditional industrial humanoid weighing 70 to 80 kilograms experiences a balance failure or a software lockup, its high-ratio planetary or cycloidal gearboxes turn its limbs into rigid steel crowbars. The kinetic energy of an uncontrolled 80 kg metallic fall is more than enough to fracture human bones or destroy residential furniture.

This safety reality is why Norway-founded and California-based 1X Technologies (backed prominently by the OpenAI Startup Fund and EQT Ventures) executed a radical mechanical departure with NEO. Instead of scaling down an industrial factory worker, 1X engineered a clean-sheet consumer biped designed around mechanical compliance, acoustic stealth, and lightweight materials. Weighing an astonishing 30 kg (66 lbs)—less than half the mass of competing commercial humanoids—NEO replaces heavy rigid reduction gearboxes with proprietary gearless, high-torque direct-drive motors linked to flexible synthetic tendon transmissions. The result is a machine that is inherently soft, mechanically backdrivable, and operating at a whisper-quiet 22 decibels, establishing the first viable architectural blueprint for safe, daily in-home human-robot coexistence.

+---------------------------------------------------------------------------------------+
| KEY TAKEAWAYS                                                                         |
| • Ultra-Lightweight Safety Profile: At just 30 kg (66 lbs), NEO eliminates the lethal  |
|   inertial impact risk of heavy 70-80 kg industrial humanoid frames.                  |
| • Gearless Revo1 Tendon Actuation: High-torque direct-drive motors eliminate rigid,    |
|   pinch-prone gearboxes, achieving 95% passive mechanical backdrivability.             |
| • Whisper-Quiet Domestic Acoustic Envelope: Operates at a measured 22 dB—quieter than |
|   a household refrigerator—banishing the loud gear whine of factory actuators.        |
| • Human-Scale 22-DoF Manipulation: Hand architecture packages 22 degrees of freedom,   |
|   driven via low-mass synthetic forearm tendons with optical tactile force feedback.  |
| • Physical Soft-Body Integration: 3D-lattice polymer impact core covered by a         |
|   removable, machine-washable knit textile suit eliminates exposed joint pinch hazards.|
| • Cognitive Intelligence Architecture: Powered by 1X's Redwood Vision-Language-Action |
|   platform accelerated locally on NVIDIA Jetson Thor edge computing silicon.           |
+---------------------------------------------------------------------------------------+

Quick Specs: 1X NEO In-Home Architectural Breakdown

The Physics of Compliance: Direct-Drive Tendons vs. Rigid Gearboxes

To understand why traditional humanoid actuation fails in the home, one must evaluate the mechanical property known as reflected inertia.

When an electric motor connects to a high-ratio gearbox (such as a 100:1 harmonic or cycloidal drive), the motor’s rotor inertia is multiplied by the square of the gear ratio ($N^2$) when felt at the output joint. If a child runs into a robot joint geared at 100:1, the joint resists movement with 10,000 times the rotor’s rotational inertia. Even if the robot’s cameras detect the child and tell the software to stop, the laws of classical mechanics dictate that the physical joint remains completely unyielding during the initial impact milliseconds. In robotics engineering, this phenomenon is responsible for severe blunt-force trauma and crushing injuries.

Reflected Inertia Comparison: The Impact Mechanics:

High-Ratio Industrial Joint (100:1 Gearbox):
[Human Impact Force] ──> [Stiff Gear Teeth] ──> [Rotor Locked by N² Reflected Inertia]
                                   │
                    (Result: Zero Give, High Impact Shock, Severe Crush Hazard)

1X NEO Gearless Tendon Joint (1:1 / Direct-Drive Ratio):
[Human Impact Force] ──> [Flexible Polymer Tendon] ──> [Ultra-Low Rotor Inertia Spins Freely]
                                   │
                    (Result: 95% Passive Mechanical Compliance, Joint Instantly Yields)

1X Technologies sidesteps this hazard entirely through its proprietary Revo1 brushless direct-drive motors and tendon-driven transmission channels.

1. Elimination of High-Ratio Gearing: Instead of relying on delicate gear teeth to amplify torque, the Revo1 motors utilize high-pole-count stators, concentrated magnetic windings, and large air-gap diameters to produce high continuous torque at a 1:1 or very low reduction ratio.
2. True 95% Backdrivability: Because there are no binding planetary gears or friction-heavy harmonic cups, the joints exhibit near-total backdrivability. If an adult, child, or pet pushes against NEO’s arm or torso, the limb simply yields physically. The motor turns smoothly in reverse, absorbing kinetic shock through mechanical compliance before any safety software or emergency stop loop even registers the event.
3. Biological Musculoskeletal Routing: In industrial bipeds, motors mount directly at the pivot points (heavy knee motors, heavy elbow motors, heavy wrist assemblies). NEO mimics biological musculoskeletal geometry. Prime movers are bundled centrally within the upper torso, pelvis, and forearms. High-tensile braided synthetic polymer tendons—similar to high-molecular-weight polyethylene fibers used in aerospace—route through sealed low-friction sheaths down the limbs to articulate the knees, ankles, and fingers. This strips rotational mass from the extremities, keeping limb inertia minimal.

The Acoustic Problem: Banishing the 65-Decibel Factory Whine

Industrial environments are notoriously loud. In a metal fabrication shop or automotive stamping bay, ambient noise easily exceeds 75 to 85 decibels. Consequently, industrial robot manufacturers do not prioritize noise attenuation. When an Atlas, Figure 02, or Unitree biped walks, the high-speed spinning of internal planetary carriers, harmonic wave generators, and cooling fans produces a distinctive high-pitched mechanical whine measuring between 60 and 70 dBA.

Introduce a 65 dB machine into a quiet suburban home, and it becomes unbearable. A typical residential living room sits at an ambient background noise level of roughly 30 to 35 dB. An industrial biped walking across tile would sound like a running vacuum cleaner or a loud lawnmower operating in the hallway.

Acoustic Signature Comparison in Residential Environments (dB Scale):
Typical Dishwasher (Modern):        44 dB  [████████████████████]
Standard Home Refrigerator:         32 dB  [██████████████]
1X NEO Measured Operational Sound:  22 dB  [██████████] (Whisper-Quiet)
Industrial Humanoid (At Walk):      68 dB  [██████████████████████████████]

NEO achieves an independently verified operational acoustic footprint of 22 dB—making it quieter than almost all modern residential refrigerators and comparable to a human whisper.

This acoustic breakthrough stems directly from its mechanical architecture:

• Zero Gear Mesh Noise: High-speed metal-on-metal or ceramic gear meshing is the primary source of high-frequency noise in electromechanical drives. Because NEO’s Revo1 direct drives eliminate high-ratio gearboxes, gear-whine harmonics are physically nonexistent.
• Polymer Tendon Damping: The flexible polymer tendon strands act as natural vibration dampers, soaking up motor micro-chatter before it can resonate through the structural skeleton.
• Fanless Conduction Cooling: NEO rejects heat into an internal aluminum heat-spreader network encased in passive airflow channels, avoiding the high-RPM cooling fans common to high-wattage computing and motor controllers.

Structural Design: 3D-Lattice Polymers and the Washable Suit

Exposed joints and sharp metal chassis are catastrophic pinch-point hazards. A traditional humanoid with exposed knee scissor-links or elbow pivots can crush a finger if a child touches the joint while it articulates.

NEO Structural Cross-Section:
[Deep Skeletal Core: High-Strength Aluminum / Carbon Torsional Spine]
  └── [Flexible Energy-Absorbing 3D-Lattice Polymer Impact Buffer]
        └── [Sealed Neoprene / Silicone Pinch-Proof Rotary Joint Boots]
              └── [Removable Machine-Washable Knit Outer Textile Suit]

NEO discards exposed mechanical linkages entirely in favor of an integrated soft-body safety matrix:

1. 3D-Lattice Polymer Skeleton: Surrounding the internal aluminum load-bearing spine is a thick, compliant exoskeleton 3D-printed from flexible lattice polymers. This cellular structure functions identically to modern athletic helmets and running shoe soles: if the robot collides with a wall or doorframe, the polymer cells buckle predictably, dissipating impact force across an expanded surface area.
2. Pinch-Proof Rotary Articulation: Every joint axis—neck, shoulders, elbows, hips, knees, and ankles—is completely encapsulated beneath sealed elastomeric boots. There are zero open gaps, exposed pulley wheels, or scissor-hinges. It is physically impossible to pinch fingers or snag pet fur in the moving joints.
3. Machine-Washable Knit Suit: The exterior aesthetic layer is not a painted plastic shell, but a durable, textured, machine-washable knit fabric suit (offered in tan, gray, and dark brown tones). This gives the machine an approachable, soft tactile feel, dampens secondary sound, and allows homeowners to unzip, wash, or swap the exterior skin when it collects household dust or stains.

Manipulation: The 22-DoF Tendon Hand Architecture

Domestic tasks demand manipulation capabilities that vastly exceed warehouse logistics. In a distribution hub, moving standardized tote boxes can be accomplished with high-pressure suction cups or rigid two-jaw parallel clamps. In the home, a robot must perform tasks like pulling a single egg from a carton, uncapping a medicine bottle, folding a soft towel, turning a round brass door knob, and threading a charging cable.

NEO Forearm and Hand Manipulator Architecture:
[Forearm Block: Array of High-Torque Micro-Actuators + Controllers]
       │
       ├── (CAN-FD Serial Bus Communication Layer)
       │
[Low-Friction Carpal Conduit Tunnel (Wrist Section)]
       │
       ├── [Tendon Channel: High-Tensile Polyethylene Filaments]
       │
[Articulated 22-DoF Hand Geometry]
       ├──> Multi-Axis Dual-Opposing Thumb Assembly
       ├──> 4 Fully Independent Digits (Multi-Joint Flexion & Lateral Spread)
       └──> Integrated High-Density Fingertip Tactile Sensor Arrays

Each hand on NEO incorporates 22 fully actuated degrees of freedom (DoF), scaling to a total of 25 DoF when factoring multi-axis wrist articulation.

• Forearm Actuation Density: To keep the hand lightweight, all driving motors are clustered inside the forearm. Ultra-thin, high-strength synthetic tendons transmit tensile forces into the fingers, enabling closing speeds of up to 8 meters per second while maintaining delicate micro-force modulation.
• Tactile Optical Skin Arrays: High-resolution tactile arrays cover the inner finger pads and palm surfaces. Rather than measuring only binary touch, these sensors quantify normal compression force, localized surface deflection, and shear slip vectors.
• Millinewton Precision: Grip force is programmatically clamped to safe thresholds (maximum continuous grip capped around 20 to 50 newtons per finger depending on object hardness), preventing the robot from crushing fragile ceramics, delicate fruit, or a human hand during physical interaction.

Operational Demonstration: Home Autonomy and Interaction

1X’s mechanical philosophy was put on display during their product launch and in-home testing demonstrations, showcasing dynamic balance, household chore execution, and compliant physical interaction:

Official 1X NEO Hardware & Chore Showcase:

Watch the platform in domestic deployment:Meet NEO: The Humanoid Home Robot from 1X

(Watch for: the fluid compliant arm movements, whisper-quiet footfalls on hard flooring, fine-motor grasping of fragile household items, and the soft-suit aesthetic integration).

Physical AI: Redwood VLA and NVIDIA Jetson Thor

To operate safely in unstructured residential spaces, NEO relies on an advanced physical AI pipeline developed in close technical coordination with OpenAI.

NEO Physical AI and Control Pipeline:
┌──────────────────────────────────────────────────────────────┐
│ HIGH-LEVEL COGNITIVE & LANGUAGE INTERACTION (10 Hz - 30 Hz)  │
│ • Multimodal LLM Voice Engine (Natural Language Understanding)│
│ • Scene Semantic Understanding & Contextual Task Graphing    │
│ • Long-Term Spatial Memory & Object State Tracking          │
└──────────────────────────────┬───────────────────────────────┘
                               │ Low-Latency Internal Interface
┌──────────────────────────────▼───────────────────────────────┐
│ REDWOOD VISION-LANGUAGE-ACTION (VLA) MODEL (50 Hz)            │
│ • Dual 8.85 MP Fisheye Camera Video Stream Processing        │
│ • Direct Pixel-to-Action End-to-End Motor Trajectory Output  │
│ • Real-Time Dynamic Obstacle & Human Trajectory Avoidance    │
└──────────────────────────────┬───────────────────────────────┘
                               │ Deterministic Low-Level Bus
┌──────────────────────────────▼───────────────────────────────┐
│ WHOLE-BODY COMPLIANT CONTROLLER (1000 Hz)                    │
│ • Direct-Drive Motor Torque Regulation & Slip Prevention     │
│ • Tendon Tension Balancing & Zero-Moment Point Gait Stability │
│ • Instant Mechanical Emergency Backdrivability Clamping      │
└──────────────────────────────────────────────────────────────┘

1. The Redwood VLA Foundation Model: Rather than programming rigid geometric rule trees for every household appliance, NEO runs 1X’s proprietary Redwood AI foundation model. Redwood combines visual scene recognition, conversational speech processing, and physical action planning into a single multimodal neural model. If a user says, “NEO, please put the sourdough bread into the pantry and close the door,” the model identifies the bread, computes compliant grasp vectors based on its soft crust, plans a collision-free arm trajectory, and regulates closing force on the pantry latch.
2. Localized NVIDIA Jetson Thor Edge Silicon: Cloud computing introduces severe latency spikes and privacy concerns. A home robot sending raw internal camera video to external cloud servers represents a major cybersecurity vulnerability. NEO processes its primary VLA models locally using the NVIDIA Jetson Thor system-on-module, delivering over 2,000 TFLOPS of FP4 physical AI compute. Vision processing, language comprehension, and balance loops execute directly on the local silicon.
3. The “Expert Mode” Teleoperation Loop: For novel, highly complex, or unmapped chores, NEO features an encrypted, opt-in Expert Mode. Trained human teleoperators wearing VR headsets can remotely assume high-level control of the robot to complete an unfamiliar task (such as sorting complex medication or cleaning an unfamiliar espresso machine). The robot records the joint telemetry, visual video stream, and tactile sensor inputs during the human demonstration, automatically using the data to fine-tune its localized autonomous policies for future execution.

Commercial Outlook: The $20,000 In-Home Consumer Reality

Deploying robots in homes demands an entirely different financial equation than industrial automation. Factories calculate payback periods based on displacing $45/hour union workers over multi-shift schedules. Consumers measure robots against the price of domestic cleaning services, private eldercare assistants, and high-end consumer electronics.

Humanoid Market Segmentation: Industrial vs. Domestic

Industrial Humanoids (Tesla Optimus, Figure 02, Atlas):
• Target Sector: Automotive, Warehousing, Heavy Logistics
• Mass Profile: 57 - 80+ kg (Heavy, Rigid Steel/Aluminum Frames)
• Actuation: High-Ratio Cycloidal / Planetary Gearboxes (Zero Backdrivability)
• Operating Environment: Controlled Factory Floors, Safety Cages, Cleanrooms
• Target Pricing: $25,000 (At High Volume) to $150,000+ (Pilots)

Domestic Humanoid (1X NEO):
• Target Sector: Private Residences, Assisted Living, Home Chores
• Mass Profile: 30 kg (Ultra-Lightweight Lattice Polymer / Fabric Suit)
• Actuation: Direct-Drive Revo1 Tendon Actuators (95% Backdrivable)
• Operating Environment: Unstructured Living Rooms, Around Children & Pets
• Commercial Model: ~$20,000 Outright Purchase or ~$499/Month Subscription

1X Technologies has structured its commercial entry with a projected $20,000 base purchase price or a flexible $499 per month consumer hardware-as-a-service (HaaS) lease.

• Automated Factory Production in Hayward: 1X established its primary manufacturing hub in Hayward, California, supplemented by research facilities in Moss, Norway. The factory is tooled to manufacture upwards of 10,000 domestic humanoids annually, using automated component manufacturing for its Revo1 motors and polymer 3D-lattice chassis structures.
• Mass Consumer Pre-Orders: The platform demonstrated immediate consumer demand, capturing more than 10,000 pre-orders within days of its public commercial unveiling.
• Target Domestic Chore Suite: Initial deployments focus on predictable domestic routines: tidying clutter, organizing laundry, setting and clearing dining tables, carrying luggage, and monitoring home security while owners travel. Over-the-air software updates continuously expand the autonomous task library as the global fleet gathers real-world training demonstrations.

1X NEO proves that the path to consumer humanoid adoption is not won by simply adding rubber pads to an industrial robot. By fundamentally replacing rigid, noisy gearboxes with gearless direct-drive motors, soft polymer structures, whisper-quiet tendon transmissions, and localized foundation AI, 1X has designed a machine that can safely and harmoniously share a home with human beings.

Explore related models and technical profiles in the Bot.to Humanoid Directory or read our direct hardware breakdown: 1X NEO vs. Tesla Optimus: Domestic Assistant vs. Factory Worker Design Philosophy.