# Menlo

## 1-line

Menlo is building Asimov, the best-of-class humanoid robot for researchers and builders.

### Why Humanoids?

We believe that humanoids will be the dominant, scale-manufactured form factor for robots:

- Humanoids can use the same tools, interfaces, and environments as humans
- Human form factor datasets are growing exponentially, through egocentric bodyworn cameras and allocentric CCTV videos
- Humans attribute greater value (and willingness to pay) to humanoid robots, due to [Anthropomorphic Bias](https://en.wikipedia.org/wiki/Anthropomorphism).

We define humanoids as a spectrum of human-form factor designs (head, 2 arms, torso, "legs") designed to operate in human-centric environments, that perform tasks that are typically done by humans.

Menlo's Asimov can be wheeled or legged, and Asimov's base has a utilitarian design that can be extended to have human-like features.

### Why Researchers & Builders?

Humanoid market is still very young, but will grow in years ahead
- Humanoid market is approximately 2,500 (2024), 7,000-10,000 humanoids (but is growing at X% CAGR) to approx $3bn by 2060
- Unitree (Hangzhou) dominates with approx. 80% of market, followed by AGIBot (Shanghai), Booster Robotics (Shanghai)

Who are buying these humanoid robots?
- Initial demand from influencers and marketing/entertainment agencies, for teleoperated humanoids as novelty performers at events
- Menlo's view is that Long-term demand requires autonomy
- Autonomy is driven by startups and researchers in university labs
- Modern humanoid robotics is still in early exploratory phase that requires hardware-software-model co-design (e.g. testing different sensor outputs to multimodal models, trying different end-effectors).

We focus on Researchers & Builders as the dominant strategy, and we build Asimov to enable this group, and be the best platform for autonomy research.

### Why Open-Source?

We believe that humanoids will be infinitely customized, and an open-source reference design GTM allows us to rapidly build a large ecosystem, with a clear path towards monetizing via proprietary components and supporting services.

- Humanoids today (e.g. Unitree) are sold as closed applicances, which throttles experimentation with vendor-in-the-loop
- Builders today want to customize Asimov, and want to work with a team that is highly responsive to building customizations for them
- Asimov's open-source reference design enables a supporting ecosystem, e.g. supply chain for replacement/repair

We envision long-term monetization around:

- High-margin proprietary components (e.g. RPUs, Sensors)
- High-margin software components (e.g. robot skills, full-body teleoperation)
- High-margin Cloud services (e.g. Simulation, Data Collection, Teleoperation)
- OEM partnerships for scale manufacturing, with certifications and licensing
- (Less certain) Dataset Collection services, or ecosystem of Dataset Collectors

Open Source Hardware allows us to engage and grow with the community.

We previously built [Jan](https://www.jan.ai/), that reached 4 million downloads and has an active, vibrant community with natural distribution and capital-efficient growth.

# Appendix

## What Menlo does:

### Menlo Asimov (Robot)

> https://asimov.menlo.ai

- Menlo's Asimov is an open-source humanoid built for researchers & builders for experimentation, accessibility and the feedback loop required to build better datasets and brains
- Asimov hardware designs are released under MIT licence, enabling anyone to build, modify, or manufacture
- Asimov can be 3D printed by researchers or purchased from certified manufacturers (Similar to OpenArm[^1])
- Asimov has a vibrant ecosystem of suppliers (e.g. OEMs, manufacturers etc, repair shops)
- Asimov has a vibrant ecosystem for researchers (e.g. egocentric datasets, models that are compatible)
- Asimov is optimized for Real-to-Sim and Sim-to-Real with Menlo's World Simulator

### Menlo Neocortex

> https://neocortex.menlo.ai

- Menlo's Neocortex is an proprietry robotic operating system that is AI-Agent and humanoid-centric (vs. the traditional ROS2 stack designed for industrial robots)
- Menlo Pi is a proprietary x86-based robot processing unit (RPU) that provides GPS and 5G connectivity, and is optimized to run Neocortex (and optionally Brain models at the edge)

### Menlo World Simulator

> https://uranus.menlo.ai

- Menlo's World Simulator lowers the cost of failure, enable rapid experimentation and infinite data generation, and expand access to people who don't have access to physical robots
- Menlo's World Simulator works with businesses and property owners to build photorealistic digital twins of real-life environments, enabling virtual proof-of-concepts of robot deployments by researchers and vendors

### Asimov Dataset

Part of https://robot.menlo.ai

- Asimov Dataset is a community-contributed dataset for humanoid learning, collected from telemetry across all Asimov robots
- Dataset includes sensor streams (camera, IMU, force, proprioception), actions (motor commands, poses), annotations (task labels, success/failure), and simulator parity data (real-world paired with simulation equivalents)
- Dataset collection leverages Neocortex's telemetry pipeline to create the feedback loop
- More robots -> more data -> better models -> more valuable robots

### Menlo Brain

Part of https://neocortex.menlo.ai

- Menlo Brain is an AI Agent system for humanoid control, using modular agent architecture (perception, reasoning, control, coordination)

### Menlo Cloud

> https://cloud.menlo.ai

- Menlo Cloud is a Robotics cloud that provides IaaS services for robotics researchers and deployments
- Menlo Cloud supports common data engines to collect robot telemetry and data for both real-world and simulated environments (e.g. Rerun.io, Prometheus/Grafana)
- Menlo Cloud provides cloud-based LLM, VLM and CV inference for Menlo Robot deployments
- Menlo Cloud powers Menlo's World Simulator, providing scalable compute for photorealistic rendering and physics engines

[^1]: https://docs.openarm.dev/purchase/