1X NEO
Home > Hardware > Humanoids > NEO
Overview
1X NEO is a consumer home humanoid robot developed by Norway-based robotics company 1X Technologies. Backed by OpenAI investment, 1X aims to create robots that can safely coexist with humans, combining their proprietary tendon-driven system with the Redwood AI VLA model to present a humanoid optimized for home environments.
| Item | Spec | Source |
|---|---|---|
| Manufacturer | 1X Technologies | [1] |
| Headquarters | Moss, Norway (relocated to Palo Alto, USA in 2024) | [3][4] |
| Height | Approximately 165-168cm (5’5”-5’6”) | [1][5] |
| Weight | Approximately 30kg (66 lbs) | [1] |
| Lifting Capacity | 68kg (150 lbs) or more (per announcement) | [1] |
| Carrying Capacity | Approximately 25kg (55 lbs) | [1] |
| Hand DoF | 22 DOF (total 75 DOF announced) | [1] |
| Movement Speed | Walking approximately 4km/h, max approximately 12km/h | [1] |
| Battery | 842Wh, approximately 4 hours operation (per announcement) | [1] |
| Charging Time | Approximately 24 minutes (fast charging, per announcement) | [1] |
| Noise Level | Approximately 22dB (per announcement) | [1] |
| Water Resistance | Hands: IP68, Body: IP44 (per announcement) | [1] |
| Drive Method | Tendon Drive | [1] |
| AI Model | Redwood AI | [2] |
| Target Price | $20,000 (monthly $499 subscription option) | [1][6] |
Note: Some specs may vary slightly by source; figures above are based on 1X official announcements.
Key Significance
NEO’s key significance is the attempt to enter the consumer home market with a humanoid. While existing humanoid robots were limited to industrial or research use, NEO was designed from the start to target ordinary homes.
New Paradigm for Home Robots
- Accessible Price: The target price of $20,000 is at the automotive level, purchasable by general households like premium home appliances
- Safety-First Design: Designed to operate safely in environments with children, pets, and elderly
- Progressive Learning: Performs basic tasks at launch, capabilities expand through software updates over time
- Human-in-the-Loop: Unknown tasks are guided remotely by 1X Expert to learn new tasks
Why Home Use?
The home environment is one of the most challenging environments in robotics:
- Unstructured Environment: Object placement changes daily
- Diverse Objects: Thousands of different items
- Human Interaction: Responding to unpredictable human behavior
- Safety Requirements: Protecting vulnerable family members
NEO directly addresses these challenges, aiming for a data flywheel effect: More NEOs operating in homes generate more data, which improves Redwood AI, and improved AI makes better NEOs.
Safety Design
NEO’s safety design philosophy is based on Intrinsic Safety. Rather than relying solely on software control, the hardware itself is designed to be safe.
Tendon Drive System
1X’s Tendon Drive System is NEO’s core safety technology [1]:
| Feature | Description |
|---|---|
| Structure | High-torque-density motors + flexible polymer tendons |
| Backdrivability | High backdrivability (approximately 95% per announcement) |
| Shock Absorption | Low impact force, smooth movement |
| Biomimicry | Mimics human muscle-tendon structure |
Why Tendon Drive?
Traditional robots have motors directly connected to joints, resulting in high rigidity. This transmits high impact force during collisions. In contrast, tendon drive:
- Natural Compliance: Tendons absorb shocks, naturally yielding during collisions
- Smooth Movement: Low jerk prevents sudden movements
- Quiet Operation: Low noise suitable for home environments
- High Torque Density: Achieves high lifting capacity while maintaining safety
Additional Safety Features
- Soft Shell: Deformable shell covered with 3D lattice polymer
- Pinch Prevention: All joints covered to prevent external access
- Situational Awareness: Environmental perception through vision and audio systems
- Remote Guidance: Human-in-the-loop approach for handling unknown situations
Version History
| Version | Release | Key Features | Source |
|---|---|---|---|
| NEO Beta | August 2024 | Initial prototype, basic bipedal walking demonstration | [5] |
| NEO Gamma | February 2025 | Enhanced dexterity, refined design, strengthened Redwood AI integration | [5] |
| NEO (Consumer) | October 2025 | Consumer version pre-orders start, US delivery scheduled for 2026 | [6] |
Note: Refer to official announcement materials for exact release dates.
Redwood AI Integration
Redwood AI is a Vision-Language-Action (VLA) model developed in-house by 1X, serving as NEO’s “brain” [2].
Technical Architecture (Per Announcement)
| Component | Description |
|---|---|
| Model Size | Transformer-based (refer to official announcement for specific parameter count) |
| Execution Environment | NEO onboard GPU (per announcement) |
| Execution Speed | Several Hz level (approximately 5Hz per announcement) |
| Input | Pre-trained language embeddings + vision tokens + proprioceptive embeddings |
| Output | Action generation through Diffusion Policy |
Key Feature: Whole-Body Control
Redwood’s biggest technical differentiator is integrated control of locomotion and manipulation:
- Most robot systems: Separate walking (locomotion) and manipulation
- Redwood: Single model controls entire body
Actions made possible:
- Bending hip and spine while reaching to pick up clothes from floor
- Balancing while supporting with one hand when opening heavy door
- Carrying objects while climbing stairs
Input Processing
[Language Embeddings] + [Vision Tokens] + [Proprioceptive Embeddings]
↓
Transformer Blocks
↓
Latent Representation Vector
↓
Diffusion Policy
↓
NEO Action OutputVoice Control
- Utilizes offboard Speech-to-Speech LLM
- Extracts user intent from conversation context
- Vectorizes commands with sentence encoder
- Inputs to Redwood model for execution
Learning and Generalization
- Data Sources: Teleoperated and autonomous episodes from EVE and NEO platforms
- Learning from Failure: Utilizes not only success but also failure data
- Robust Generalization: Can pick up unseen objects from various positions
1X Technologies Company Information
| Item | Details |
|---|---|
| Founded | 2014 (formerly Halodi Robotics) |
| Founder/CEO | Bernt Bornich (Bernt Bornich) |
| Headquarters | Founded in Moss, Norway, later relocated to Palo Alto, USA |
| Key Investors | OpenAI, Tiger Global, EQT Ventures, Samsung NEXT |
| Total Investment | Over $137M (as of 2024) |
Investment History
| Time | Round | Amount | Key Investors | Source |
|---|---|---|---|---|
| March 2023 | Series A2 | $23.5M | OpenAI Startup Fund (lead), Tiger Global | [4] |
| January 2024 | Series B | $100M | EQT Ventures, Samsung NEXT | [4] |
Partnership with OpenAI
1X contacted OpenAI in the summer of 2022, before the release of ChatGPT. They subsequently integrated OpenAI’s language models and embodied learning models into robots to build a system that understands and learns from natural language commands [4].