![[2602.17174] Continual uncertainty learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.17174] Continual uncertainty learning
Summary
The paper presents a novel framework for continual uncertainty learning in robust control of nonlinear dynamical systems, addressing challenges in deep reinforcement learning and sim-to-real transfer.
Why It Matters
This research is significant as it tackles the persistent issue of managing multiple uncertainties in control systems, which is crucial for advancing applications in robotics and automotive engineering. By improving learning efficiency and policy robustness, it has the potential to enhance real-world implementations of machine learning in complex environments.
Key Takeaways
- Introduces a curriculum-based continual learning framework for robust control.
- Decomposes complex control problems into manageable tasks to handle uncertainties.
- Incorporates a model-based controller to enhance learning efficiency.
- Demonstrates practical application in designing an active vibration controller for automotive powertrains.
- Achieves successful sim-to-real transfer, addressing the sim-to-real gap.
Computer Science > Machine Learning arXiv:2602.17174 (cs) [Submitted on 19 Feb 2026] Title:Continual uncertainty learning Authors:Heisei Yonezawa, Ansei Yonezawa, Itsuro Kajiwara View a PDF of the paper titled Continual uncertainty learning, by Heisei Yonezawa and 2 other authors View PDF Abstract:Robust control of mechanical systems with multiple uncertainties remains a fundamental challenge, particularly when nonlinear dynamics and operating-condition variations are intricately intertwined. While deep reinforcement learning (DRL) combined with domain randomization has shown promise in mitigating the sim-to-real gap, simultaneously handling all sources of uncertainty often leads to sub-optimal policies and poor learning efficiency. This study formulates a new curriculum-based continual learning framework for robust control problems involving nonlinear dynamical systems in which multiple sources of uncertainty are simultaneously superimposed. The key idea is to decompose a complex control problem with multiple uncertainties into a sequence of continual learning tasks, in which strategies for handling each uncertainty are acquired sequentially. The original system is extended into a finite set of plants whose dynamic uncertainties are gradually expanded and diversified as learning progresses. The policy is stably updated across the entire plant sets associated with tasks defined by different uncertainty configurations without catastrophic forgetting. To ensure learning effi...
