articleFeb 11, 2026
Supply Chain

Beyond the Hype: Finding the Sweet Spot in Robotics

The AI Inflection Point in Robotics

 

In 1962, the world’s first industrial robot, Unimate, was deployed on a General Motors assembly line, ushering in a new era of automation.1 For the sixty years since, that promise remained largely unfulfilled for most of the economy. Early robotics were rigid, capital-intensive, and required specialized expertise. Their benefits were limited to standardized processes accessible only to large enterprises. That is finally changing. This article explores the technological breakthroughs driving today’s robotics resurgence and identifies where we see the most compelling opportunities.

Beyond the Hype Highlights

The Breakthrough

Robotics is evolving from rigid machines to embodied intelligence powered by foundation models, with world models enabling robots to simulate and plan before acting.

The Tension

The robotics market is split between high generalization humanoids, which are capital-intensive and slow to commercialize, and high-performance systems for constrained environments.

The Investment Thesis

A performance-first approach like DYNA Robotics enables faster deployment, early revenue, and proprietary data collection to build a durable moat for future generalization.

The Competitive Edge

Companies prioritizing commercial reliability gain accelerated learning advantage, keeping systems competitive even after humanoids overcome cost and safety hurdles.


Prior to 2022, robotic control was rooted in classical autonomy policies using a modular sense–plan–act (SPA) architecture, that is, perception, planning & action were each enabled by a separate system. This architecture offered greater interpretability (easier to debug) and higher data efficiency, but it struggled to generalize. Each new task required engineers to write new rules or redesign subsystems, limiting scalability.

 

The turning point traces back to 2017, when Vaswani et al.  introduced the Transformer architecture in their landmark paper, “Attention Is All You Need.” This architecture unlocked a model’s ability to weigh the contextual importance of different pieces of data, enabling large language models to process meaning rather than sequence alone. This context reasoning ability is now being applied to robotics. Transformers are powering end-to-end large-scale robotic foundation models that unify perception, control, and reasoning, marking a leap toward embodied intelligence.

 

 

Most current systems are Vision-Language-Action (VLA) models, but they remain reactive, responding moment-to-moment without foresight. These limitations led to the development of world models, creating architectures that fuse perception with learned simulations of physical environments. A world model essentially allows a robot to “think ahead,” plan through simulated experience, and execute long-horizon tasks more efficiently. This concept was dramatically scaled in early 2025 with the launch of the NVIDIA Cosmos World Foundation Model Platform. Designed as a "synthetic data generation and augmentation platform", that integrates with tools like Omniverse, Cosmos uses foundation models to generate vast amounts of photoreal, physics-accurate data for training and evaluating robot policies. This technology provides a crucial solution to help accelerate development across all Physical AI systems.

 

 

This combination of capabilities has fueled rapid commercialization. Agility Robotics’ Digit is leveraging a whole-body control foundation model to enable coordinated motion, walking, and object manipulation. These milestones confirm that foundation models are redefining robot autonomy, perception, and control into commercially viable solutions.

Mapping the robotics landscape


To understand where robotics is headed, we look at two key dimensions: generalization and performance. Together, they map the paths from research breakthrough to commercial scale.

 

Defining generalization

 

As defined by Prof. Deepak Pathak of Carnegie Mellon University, a leading robotics researcher and founder of Skild AI:

 

“Generalization is the robot’s ability to perform a wide variety of tasks and in a wide variety of environments using a single control policy.”

 

In practical terms, a cleaning robot has low generalization because it performs fixed, repetitive tasks in controlled environments. By contrast, a humanoid robot has high generalization, adapting across diverse settings and task types with a single control policy.

 

Defining performance

 

Performance is measured by how effectively a robot executes its intended tasks, which we evaluate across three KPI’s:

 

  1. Human Intervention Rate — how often an operator must step in.
  2. Comparative Human Speed — task completion time relative to human workers.
  3. Task Accuracy — how precisely and consistently the robot performs.

 

High-performing systems minimize interventions, operate at or above human speed, and maintain precision over extended periods. In contrast, even advanced robots with strong generalization can struggle if their performance isn’t commercially viable at scale.

 

Supply Chain, Logistics & Advanced Manufacturing Annual Report | 2025

Supply Chain Annual Report | 2025
Download Now

 

The Categories


We now assess these categories based on their position within the performance vs. generalization matrix:

 

 


Specialized robots
(low generalization, high performance)

 


These robots represent the peak of performance but the limit of generalization. Classic examples include painting, welding, and pick-and-place arms, which dominate industrial settings because of their precision and reliability in repetitive tasks. Among these, Autonomous Mobile Robots (AMRs) have become the most widely adopted, with over $1.2 million active units worldwide. However, foundation-model-driven autonomy is transforming them into situational collaborators — robots that can handle dynamic obstacles, understand human instructions, and adapt in real time to “brownfield” conditions. This evolution underscores a major shift in robotics: from pre-scripted automation toward contextual autonomy, where robots can make decisions rather than just execute them.

 

 

General-purpose dexterous manipulation robots (medium generalization, high performance)

 


These systems represent a crucial, performance-first approach to the market. Their goal is to deliver commercial-grade dexterity and reliability within a fixed workspace before expanding outward. Startups like DYNA Robotics are defining this category, specializing in high-precision manipulation tasks such as packaging, assembly, and quality inspection. By focusing on constrained environments, they achieve real-world deployment faster. For example, the DYNA-1 achieved a 99.4% autonomous success rate over 24 hours of continuous operation in revenue-generating pilots.  These systems demonstrate that narrow focus can still produce broad economic impact — especially when paired with foundation models that can generalize dexterous control over time.

 

 

GET IN TOUCH

Ready to start your innovation journey?


Humanoid robots (high generalization, developing performance)


This category focuses on high mobility and adaptability in human-centric environments, resulting in high generalization but often with performance that is still in development. Companies such as Figure AI and Apptronik are pursuing a generalization-first strategy, prioritizing locomotion, balance, and adaptability in human-centric environments. Despite enormous fundraising, Figure AI raised over $1 billion in Series C funding at a $39 billion valuation in 2025, but these humanoids remain primarily in pilot and demonstration phases.

 

The challenge is structural: chasing a “human-equal” robot isn’t feasible today. Effective commercial rollout is currently slowed by hardware limitations, energy constraints, and safety certification hurdles. Crucially, the AI layer struggles to handle the high degrees of freedom (DoF) required by a true humanoid. This mismatch often forces design compromises, such as mechanical simplification in models like later Figure prototypes, revealing that while competitors iterate toward generalization via proven, deployable systems, the pure humanoid approach is trying to leap to the finish line without the foundational AI to sustain it.

 

The core thesis


As investors, we focus on companies that prioritize performance because that is what the industry demands first. While it’s tempting to chase the vision of pure generalization, the real challenge is delivering reliable function in complex, varied environments today.

 

The performance-first approach of companies like DYNA Robotics and Maven Robotics strikes a pragmatic balance, validated by the initial success of Agility Robotics. Agility’s hybrid model demonstrated that coupling a generalist platform with a performance focus on specific, simple tasks gains early market traction, successfully avoiding the pitfall of trying to do everything at once.

 

By prioritizing systems capable of commercial-grade reliability today, these companies achieve several critical advantages:

 

  1. Faster Deployment: They overcome the hurdles of reliability and safety validation sooner.


 

  1. Rapid Data Collection: Getting robots into the field enables them to immediately collect the massive, diverse, and messy operational datasets that are essential for training the next generation of foundation models.


 

  1. Accelerated Learning: This commercial velocity and rapid learning is a competitive advantage as critical as the technology itself, sharpening go-to-market strategies and accelerating generalization via real-world feedback.

Resilience as an Investment


The inherent resilience of the performance-first approach makes it a safer investment. Even when true humanoids reach parity with human capability, there will be no economic justification to deploy a highly complex, expensive, generalist robot for tasks that do not require high generalization (such as simple tote recycling). The specialized and dexterous manipulation systems will retain their market share due to their superior cost-to-performance ratio in their constrained niches.

 

 

In short, performance-first companies learn faster, deploy sooner, and capture data that forms the backbone of future generalization. We believe this approach is the most effective play in the mid-term until pure humanoids overcome the significant hurdles of cost, energy, and safety required for effective, mass-scale deployment.

 

PLUG AND PLAY AROUND THE WORLD

Discover our global locations

We have a community of experts all over the world. We work as a bridge between Silicon Valley and the rest of the world, making innovation open to anyone, anywhere.

Sign up for our global newsletter