3 results about "Merit function" patented technology

The application provides a dynamic memory enhancement method and system for a visual-language-action model and a storage medium, and solves the problems of short-sighted memory, memory pollution and inability to dynamically adjust the fusion weight of the existing VLA model in a long-distance task. The application comprises: constructing a working memory and retrieving historical memory from a perception-cognition-reward memory bank; using a gating network with trainable parameters to adaptively weight and fuse the current memory and the historical memory to generate an enhanced working memory; using a progress evaluator to calculate a task progress difference value based on the enhanced working memory to generate a single-step dense reward; using the reward as a merit function to perform online policy optimization on the gating network parameters; and when the memory bank is full, merging adjacent entries based on the weighted indicators of feature similarity and reward similarity; and finally outputting a motion sequence by a motion expert network. The application can be used for robot long-time sequence task learning and adaptive control in a dynamic environment.

A wavefront correction system and method with evaluation function switching and derivative-free trust region (RDR) is disclosed. This invention relates to the field of optimization and control technology for wavefront-free adaptive optics systems, specifically to a wavefront correction system and method with evaluation function switching and derivative-free trust region (RDR) optimization. The invention organically integrates a phased evaluation function strategy with a model-based derivative-free trust region optimization algorithm, achieving seamless switching of the evaluation function through observation data reuse. The method includes the following steps: simultaneously defining a first evaluation function and a second evaluation function; performing first-stage derivative-free trust region wavefront correction based on the evaluation function, while simultaneously constructing an observation data reuse buffer; switching to second-stage derivative-free trust region wavefront correction; selecting historical sampling points from the observation data reuse buffer, performing second-stage derivative-free trust region wavefront correction based on the observation data, and obtaining the optimal sampling point; updating the wavefront correction value based on the optimal sampling point.

The invention discloses a composite evaluation function stray light optimization method based on dual-mode ray tracing, and the method comprises the steps: 1, building a light source model and an optical-mechanical structure model; 2, according to light source information and optical system matrix information, sequence optical tracing is carried out, and a loss1 value is obtained according to a result; step 3, performing non-sequence optical tracing according to the light source information and the optical system matrix information, and obtaining a loss2 value according to a result; and step 4, optimizing the optical system by using a differential genetic algorithm in a dynamic weight distribution mode to obtain an optimized optical system. Compared with a traditional optimization mode, the optimization mode provided by the invention considers the system imaging quality while optimizing the stray light level, and meanwhile, the non-sequence tracing efficiency is improved by using the matrix, so that the time cost in the optimization process is reduced. The method is of great significance to stray light optimization of an optical system.