Juno: Memo

The robotics landscape is dynamic and increasingly crowded, particularly with the recent surge in humanoid robotics development. Amidst this activity, it’s crucial to articulate Nyro’s distinct vision and the strategic rationale behind our focus. This memo clarifies our chosen path: developing teleoperated bipedal humanoid robots designed for real-world deployment.

Exhibit A: Mapping the Robotics Deployment Landscape

To understand Nyro’s unique position, it’s helpful to categorize current industry efforts based on how companies deploy or aim to deploy robots in the wild, focusing on locomotion (Humanoid vs. Wheeled) and control (Teleoperated vs. Autonomous).

	Bi-manual Wheeled Robots	Humanoid Robots
Fully Autonomous	Physical Intelligence, Amazon Robotics, Clearpath, Botco Constrained by terrain and true generalizability	Tesla, Figure, Boston Dynamics, Agility, Apptronik, 1X, Foundation, The Humanoid, Apple, Meta, OpenAI Significant autonomy challenges remain
Teleoperated	Watney Robotics, Tangible Bot Limited by environment accessibility.	Nyro Currently Underserved Area

Definitions:

• Teleoperated: Partial teleoperation entails robots having parts of their bodies controlled by a remote human operator. For careful object manipulation tasks, this means systems for mirroring a human’s upper-body and dextrous hand movements. As a result, the teleoperator needs high-fidelity multimodal feedback regarding the robot’s state, including visual, haptic, and audio feedback.
• Fully Autonomous: Robots making their own decisions based on sensor input and internal models. Faces challenges in generalizability and handling long-horizon, complex tasks (i.e. longer than say picking up and putting down a kitchen sponge). Ultimately, based on learnings from the autonomous vehicle (AV) industry, this will not be solvable without vastly more diverse data, which these companies lack. Other robotics companies do not collect data in the wild, and their autonomous systems are therefore restricted to lab environments. It takes lots of real-world data to achieve autonomy, for example Tesla’s FSD remains level 2 and Waymo continues to only operate in small, geofenced, urban environments.
• Humanoid: Robots with two legs and two arms, designed for human-like mobility, allowing climbing and proper interfacing with the current human ecosystem. Many mechanical engineers have iterated upon varying limb counts, whether adding more arms or legs. Although additional legs and arms would theoretically allow for more performance in creating leverage for traversing difficult terrains and manipulating complex objects, the marginal utility decreases sharply with each new limb over two, causing issues in stability with reduced symmetry. Further, in reducing number of legs or arms, we conversely lose the ability to respond to disturbances or manipulate common human tools. As a result, maintaining the humanoid form factor is paramount to engaging effectively with unstructured environments.
• Bi-Manual Wheeled: Robots using wheels for locomotion, typically faster on smooth surfaces but limited by obstacles and elevation changes. Companies engaging with this approach are often restricted to very structured and indoors environments, which again, results in a long-term failure to collect generalizable and diverse data due to a lack of real-world exploration.

There is a reason why no one else is taking this approach. It’s impossible to exactly control the lower body of a bipedal robot via direct teleoperation. Transferring a human’s stabilization and balance to the legs of a humanoid isn’t possible due to the difference between our forms: while a humanoid robot has around 50 electric actuators that move the joints, humans have 630 highly specialised muscles. As a result, our approach involves shared autonomy — we understand at a high level where a human wants to move from their motions on our omnidirectional treadmill, and utilise our purpose-trained neural-network based controller to command the robot to move its legs as desired. Throughout this process, the human continues to teleoperate the robot’s upper body, manipulating objects in the robot’s environment while the robot’s legs automatically stabilise. As a result, our robots are able to interface with unstructured environments — regardless of terrain, object types, or intelligence required for the task.

Why Teleoperated Humanoid Robots? The Nyro Rationale

While the ultimate goal for many is full autonomy, we believe a crucial intermediate – and potentially enduring – step is being overlooked: effective teleoperation, particularly in a bipedal form.

1. Leveraging Human Intelligence Now: We reject the notion of waiting indefinitely for generalized AI capable of handling unpredictable, real-world tasks. Teleoperation and shared autonomy allows us to deploy robots now, utilising the unparalleled adaptability and problem-solving skills of human intelligence. It allows us to tackle complex, long-horizon tasks where current autonomy falls short.
2. The Generalizability of the Human Form: Wheeled robots have their place, but their place is predominantly flat, predictable, and structured. They simply cannot provide the general-purpose mobility needed for the truly unstructured environments humans inhabit. Their inherent limitations – requiring ramps, struggling with debris, needing specific clearances – confine them operationally and, crucially, limit the diversity of data they gather. If the goal is generalized autonomy, this constrained data is insufficient.

Nyro chooses the humanoid form because the world is designed for it. It's the most generalisable hardware for navigating complex, human-centric spaces. While harder to engineer, we tackle this challenge head-on because it's essential for our long-term vision. Pairing this adaptable form with human intelligence via teleoperation lets us deploy now into the messy reality of unstructured tasks. This immediate, real-world deployment provides the unique, diverse humanoid data needed to train truly intelligent systems – data that simpler, environmentally-constrained platforms will never access.

Companies that over-index on wheels end up building technology, then struggle to find use cases that truly justify the high cost of hardware or even the cost of teleoperators. We see these companies’ failure within GTM and aim to prioritise realistic deployment, not one-off demos.

Exhibit B: Nyro's Strategic Technical Bets

Developing shared autonomy and reliable systems for long-term remote teleoperation requires significant technical investments that other robotics companies do not intend to take. Here are some areas where we’ll have a decisive advantage:

1. Robust Bipedal Locomotion Under Upper-Body Disturbance: A teleoperated upper body introduces unpredictable dynamics to the lower body. Enabling stable walking and manoeuvring while the operator performs complex tasks is a major hurdle. Grounded in our past research at MIT, Stanford, Edinburgh and engineering experiences from Tesla and K-Scale Robotics we are making a concentrated technical bet on deep reinforcement learning (DRL) approaches to develop a highly robust, adaptive whole-body controller capable of handling this complexity.
2. Ultra-Low Latency & Secure Teleoperation: Foundational to effective remote control. The operator needs a seamless, real-time connection to the robot’s senses (especially vision) and the ability to translate their movements precisely. This involves significant challenges in networking and cybersecurity, which we are directly addressing through secure peer-to-peer connectivity and specialised codecs for relevant VR video.
3. Advanced Exoskeleton Interfaces: For high-precision tasks, particularly in high-risk zones, standard VR hand-tracking-based methods are insufficient in their sample frequency, tracking accuracy, and ability to handle occlusions. We see exoskeletons, force-feedback suits, and wearables as a key differentiator, offering more intuitive and reliable control and paving the way for haptic feedback systems, enabling more dexterous manipulation than ever before.
4. Ergonomic Operator System (VR/AR + Training): Our commitment extends to the human operator. We invest heavily in an ergonomic and intuitive VR/AR interface (UI/UX). Recognising that robot agility can induce operator discomfort, we are strategically developing cybersickness mitigation technologies as a core part of the user experience, advised by a pioneering VFX veteran with decades of real-time VR/XR experience. This focus ensures operators can comfortably maintain control and focus during complex, dynamic tasks over extended shifts, maximising the real-world utility of our system.

Like OpenAI’s focused bet on scaling transformers led to ChatGPT’s breakthrough, Nyro is concentrating its resources on the specific, interconnected challenges of deploying humanoids in the wild. This focused approach, addressing the unique demands of teleoperated legged systems, will allow us to dominate. We focus on the hard problem — imagine if ChatGPT restricted itself as wheeled robots do. We would be nowhere near where we are today.

Verticalization as a Strategic Data Engine

Further, our go-to-market strategy follows as another technical bet. We want to collect data at scale to be able to train models and eventually reach full autonomy. The only way to get exposed to the data across our target verticals is by operationalising: deploying our robots in the real-world. Being first to do so, Nyro will become synonymous with humanoid deployment in our target industries.

By initially targeting high-value, high-risk verticals — hazardous materials handling and defense — we address immediate needs where human presence is dangerous or impossible, justifying the current system costs and the humanoid form-factor. Achieving dominance within these niches provides distinct advantages over lab-bound or non-practical approaches:

1. Targeted Real-World Data: Leading a niche yields concentrated, high-quality data directly from valuable, unstructured tasks performed in real deployment environments instead of mockup lab environments. This is the essential fuel that current autonomous efforts lack.
2. Pragmatic Path to Autonomy: The data flywheel (deploy → collect → refine) allows us to build and deploy autonomous capabilities incrementally. By exposing our hardware and software to actual real-world use cases and working closely with end-users in hazardous materials and defense, we will have the most comprehensive understanding of market needs and will be able to tailor our solution to meet requirements as first movers. By closely collaborating with end-users and delivering pilots before anyone else, we will establish the indispensable trust required to succeed in our target verticals.
3. Market Expansion: As our systems mature and hardware costs decrease, this proven capability, rich real-world data foundation, and established reputation will enable strategic expansion into broader markets. As we collect data, we will continually improve our autonomous systems, incorporating feedback from real-world success signals within that vertical rather than simply imagining use cases like other companies. At end-state, through real-world deployment, we can realistically achieve full autonomy.

This verticalization strategy allows us to concentrate resources, generate revenue, and build an unparalleled dataset from a position of market leadership, allowing us to train truly intelligent networks and deploy models incrementally. As a result, we position Nyro towards being the leading contender for creating humanoids for real-world, scalable deployment.

Summary

Nyro is forging a distinct path based on a pragmatic assessment of current technological capabilities and future needs. As dropouts from leading labs in humanoid robotics and robotics AI research, we believe we have everything to succeed in the relentless pursuit of our vision. By strategically combining the strengths of human dexterity, intelligence, and the adaptability of the bipedal form, enabled by targeted deep-tech investments and a focused verticalization strategy, we aim to unlock near-term value in unstructured environments. This approach allows us to generate revenue early and gather essential real-world data, driving iterative improvements and paving the way towards full autonomy.