Robots on the rise

By Markus Schulte-Huermann | Strategie-Team

Robots on the rise

How we master epistemic uncertainty and Knodge.eu helps us to do so

The world of robotics is developing rapidly. Robots are no longer science fiction, but an increasingly integral part of our everyday lives and our industry. However, with the increasing complexity and autonomy of these systems, the need to overcome their "knowledge gaps" the so-called epistemic uncertainty , to understand and manage them. In this blog post, we take a look at current developments in the robotics market, the solutions to epistemic uncertainty and how platforms like Knodge.eu can play a crucial role.

The market for humanoid robots: Who are the players?

The market for humanoid robots is booming and is expected to grow to over $38 billion by 2035, with the cost per unit potentially falling below $20,000 The humanoid robot market 2025/2026: Everything you need to know. A large number of companies worldwide are driving this development:

These robots are designed to operate in complex, dynamic environments, which raises the question of their reliability and safety, especially when they encounter unknown situations.

Epistemic uncertainty: what is it and how is it addressed?

Epistemic uncertainty describes the knowledge gaps or inaccuracies in a robot's models that could, in principle, be reduced by more data or a better model. It is the uncertainty that arises when the robot "does not know what it does not know" An overview of uncertainty representation and quantification in neural networks. For robotics, recognizing and managing this uncertainty is critical for safety, robustness and human trust International Report on AI Safety 2026, The Humanoid Robot Market 2025/2026: Everything You Need to Know.

Various approaches are being pursued to overcome this challenge:

  1. Uncertainty quantification in AI models:
    • Bayesian neural networks (BNNs):** These place probability distributions over their weights in order to capture the uncertainty directly in the model parameters.
      Deep ensembles: Several independent models are trained here. The spread of their predictions serves as a measure of uncertainty An overview of uncertainty representation and quantification in neural networks.
      Monte Carlo Dropout: An efficient method that also uses dropout during inference to obtain a distribution of predictions and thus a measure of uncertainty.

2 Causal world models: Robots develop internal models of their environment to understand causality and predict future states. Systems such as LingBot-VA combine video world modeling and speech pre-training to "understand" the effects of actions and thus reduce epistemic uncertainty in control Causal World Model for Robot Control.

3 Continuous and adaptive learning: Robots must be able to dynamically adapt their knowledge base. Approaches such as nested learning and the dynamic adaptation of retrieval augmented generation (RAG) architectures enable robots to continuously update their models and close knowledge gaps, especially in unknown situations [Portal Knowledge Base: RAG Nested Learning](https://knodge.eu/portal/robotic?tab=documents&search=portal-memory-RAG Nested Learning-2026-01-10.pdf).

4 Recognition of out-of-distribution (OOD) situations: Robots need to recognize when they encounter situations that they have never seen in their training data. In such cases, epistemic uncertainty is at its highest, and the robot must act with appropriate caution or request human assistance.

Knodge.eu: A European partner for safe and knowledge-based robotics

This is where Knodge.eu comes into play. As a European knowledge hub, Knodge.eu offers decisive advantages to support robotic systems in coping with epistemic uncertainty:

The future of robotics is closely linked to the ability to manage uncertainty. By combining advanced AI methods and a secure, knowledge-based platform such as Knodge.eu, we can develop robots that are not only capable, but also safe, reliable and trustworthy.

Bibliography: