Dance choreography assistance system by hierarchical reinforcement learning and its real-time interaction optimization method
In view of the shortcomings of the current automatic dance generation system in terms of long-term structural consistency and users' immediate intervention ability, a set of dance choreography auxiliary framework integrating hierarchical reinforcement learning (HRL) is constructed. The system establishes a two-level decision-making framework. The upper strategy is responsible for analyzing the emotional dimension and melody transition of music, and planning the choreography logic of dance segments from the macro level, while the lower strategy focuses on the continuous generation of micro-body movements to ensure the precise alignment of posture and rhythm of each frame. In order to enhance the controllability of the creative process, the real-time interactive optimization (RTIO) method is introduced, and through the human-in-the-loop (HITL) mechanism, the creator is allowed to adjust and correct the action trajectory or artistic style in the process of algorithm generation. In the research process, FineDance, a high-quality professional data set containing various dance styles, is used for model training and parameter optimization. The results show that after the HRL framework combined with RTIO, the generation quality and control ability are obviously improved, the Fréchet Inception Distance (FID) is reduced to 16.57, and the beat hit rate is increased to 0.879. The performance of action coherence, structural stability and music matching degree is more balanced. The study shows that this method can give consideration to both long-time scheduling and real-time controllable editing in dance generation, which provides more practical technical support for the intelligent choreography system to move towards real creative scenes.
Authors
- Tang Hua (ORCID: https://orcid.org/0000-0002-1281-7865)
Institutions
- Shandong Academy of Sciences (CN)
- Shandong University of Arts (CN)
Publication Details
- Journal
- Discover Artificial Intelligence
- Published
- 2026-11-17
- DOI
- https://doi.org/10.1007/s44163-026-02737-4
- Primary Topic
- Human Motion and Animation
- Type
- article
- Field-Weighted Citation Impact
- 0.00