Joint optimization method and device, computer device and storage medium
Patent Information
- Application Number
- CN202610201777.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-11
- Publication Date
- 2026-06-09
AI Technical Summary
Existing separate optimization methods make it impossible to predict the optimal control strategy for the morphology design of magnetically driven robots. During the control phase, they can only passively adapt to the fixed morphology design, making it difficult to achieve global optimal performance. Furthermore, traditional morphology optimization relies on gradient-free evolutionary algorithms, which are computationally expensive and make it difficult to use reward signals during the control process to guide morphology improvement.
A joint optimization environment model is constructed, which models the morphological design and control optimization of the magnetically driven robot as a two-stage sequential decision-making process. The proximal policy optimization algorithm is used to perform end-to-end joint training of the two network agents. The parameter updates of the design policy network are guided by the aggregated rewards of multiple control evaluations, so as to realize the co-evolution of the design policy network and the control policy network under a unified optimization framework.
It significantly improves the overall task performance and optimization efficiency of magnetically driven robots. The magnetization configuration parameters can be adjusted in a directional manner based on the feedback of the magnetic field control signal. The control strategy network can adapt to changes in magnetization configuration in real time, avoiding local optima and reducing computational costs.
Smart Images
Figure REF-OBJ-1770815390627-000029 
Figure REF-OBJ-1770815390627-000030 
Figure REF-OBJ-1770815390627-000031