GIFT: Global stabilisation via Intrinsic Fine Tuning

2026-04-25

AI Summary

Deep reinforcement learning demonstrates strong performance in complex continuous control tasks but often lacks global stability due to high sensitivity to initial conditions, limiting its applicability in real-world systems. To address this challenge, this work proposes GIFT, a general training framework that, for the first time, explicitly incorporates global stability as an optimization objective during post-training. By introducing a tailored stability-oriented reward function and an intrinsic fine-tuning mechanism, GIFT significantly enhances policy robustness while preserving the original high task performance. Experimental results across multiple continuous control environments show that GIFT effectively improves global stability without compromising task fidelity, thereby substantially increasing the practical viability of deep reinforcement learning in real-world control systems.

Technology Category

• Machine Learning: Reinforcement Learning
• Search and Optimization: Learning to Search
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Deep reinforcement learning policies achieve strong performance in complex continuous control environments with nonlinear contact forces. However, these policies often produce chaotic state dynamics, with trivially small changes to the initial conditions significantly impacting the long-term behaviour of the control system. This high sensitivity to initial conditions limits the application of Deep RL to real-world control systems where performance and stability guarantees are often required. To address this issue, we propose Global stabilisation via Intrinsic Fine Tuning (GIFT), a general-purpose training framework which directly optimises the global stability of existing high-performing deep RL policies using a custom reward function. We demonstrate that GIFT increase the stability of the control interaction while maintaining comparable task performance, thereby improving the suitability of deep RL policies for real-world control systems.

Problem

• deep reinforcement learning
• global stability
• chaotic dynamics
• sensitivity to initial conditions
• real-world control systems

Innovation

• global stability
• intrinsic fine-tuning
• deep reinforcement learning
• continuous control
• chaotic dynamics