1X Technologies
1X Technologies, NEO Humanoid, and the World Model Approach
Overview
1X Technologies is a Norway-based humanoid robotics startup developing the android robots EVE and NEO. The company differentiates itself through the Redwood AI model and an automated evaluation/training system using World Models.
| Item | Details |
|---|
| Headquarters | Palo Alto, CA (HQ), Moss, Norway (Manufacturing) |
| Founded | May 2014 (formerly Halodi Robotics) |
| CEO | Bernt Øivind Børnich |
| Funding | $126M+ (as of Series B) |
| Investors | OpenAI, EQT Ventures, Tiger Global, Samsung NEXT |
Robot Lineup
EVE
| Item | Specs |
|---|
| Type | Wheel-based humanoid |
| Purpose | Security, facility management |
| Deployment | ADT security partnership |
| Features | 24-hour patrol capable |
NEO (2024)
| Item | Specs |
|---|
| Type | Bipedal humanoid |
| Height | 165cm |
| Weight | 30kg |
| Carrying Capacity | 25kg |
| Speed | Walking 4km/h, Max 12km/h |
| Purpose | Domestic use |
NEO Gamma (February 2025)
- Advanced version of NEO
- Improved dexterity (22 DOF per hand, 75 DOF total)
- Natural human gait (reinforcement learning based)
- 10x improved hardware reliability
- 22dB noise level (refrigerator quiet)
World Model Approach
Core Idea
1X automates robot evaluation through a Learned Simulator:
[Real Robot Data] ──Training──→ [World Model (Simulator)]
│
▼
[Automated Evaluation/Testing]
│
▼
[Policy Improvement]
Advantages
| Traditional Approach | 1X World Model |
|---|
| Manual testing required | Automated evaluation |
| Sim-to-Real gap | Learned from real data |
| Limited scenarios | Infinite scenario generation |
Research & Papers
- “1X World Model” (2024): Video-based world model
- Simulator trained on real robot data
- Used for policy evaluation and planning
World Model as Evaluator
The World Model serves beyond simulation as a Policy Evaluator:
| Use Case | Description |
|---|
| Automated Evaluation | Test new policies without physical robot |
| Action Optimization | Simulate “what happens if I take this action?” |
| Test-time Planning | Search for optimal action sequences before execution |
Similar to recent research like DexWM: test-time action optimization through World Model without policy training
Data Collection
Large-Scale Demo Collection
- Operating a team of human teleoperators
- Data collection across diverse environments
- End-to-End neural network training
Embodied Data
Simulation Data vs Real Data
1X Perspective: "Real data is most valuable"
→ Maximize utilization through World Model
Business Strategy
Consumer Robotics Market
| Timeline | Plan |
|---|
| August 2024 | NEO Beta unveiled |
| February 2025 | NEO Gamma released |
| October 2025 | NEO pre-orders started ($20,000 or $499/month) |
| 2026 | Initial US home deployment |
Security Market (EVE)
- Partnership with ADT
- Facility security patrols
- 24/7 unmanned surveillance
Differentiators
vs Tesla Optimus
| Factor | 1X | Tesla |
|---|
| Weight | 30kg (NEO) | 55-57kg |
| Target Price | $20,000 | ~$25,000 |
| Data Access | Dedicated collection | Own factories |
| AI Approach | World Model | End-to-End |
| Factor | 1X | Figure |
|---|
| Target Market | Consumer-first | Industrial-first |
| Robot Weight | Lightweight | Standard |
| Approach | World Model | VLA (Helix) |
References
See Also