The present application relates to a method, device, and electronic equipment for generating generative agents in three-dimensional space. It includes constructing a three-dimensional active scene based on preset scene elements and establishing a time system for the active scene; importing initial virtual objects and setting initial virtual object parameters; and generating a generative agent model based on the initial virtual object parameters. The method further involves training the generative agent model with a generative adversarial network (GAN) to optimize the model parameters, thereby generating a high-quality generative agent. The invention provides a solution for creating generative agents in three-dimensional space, which can be applied to various fields such as virtual reality, gaming, and animation.
背景技术
生成式智能体是一种利用生成模型来模拟可信人类行为的智能体,是近几年新兴的一种智能体构建方案,它的出现依托于大语言模型的兴起。以往的智能体的底层设计逻辑多为有限状态机或者AI(Artificial Intelligence,人工智能)行为树,这种智能体的编码难度高,且行动方式固定,难以应对各种复杂的环境以及任务。而生成式智能体可以利用大语言模型优秀的文字处理能力,来根据环境以及自身状态变化自动的执行各种动作,行动方式更加灵活多变。
相关技术中,生成式智能体的产品大多是2D(Two-Dimensional,二维)而非3D(three-dimensional,三维)。二维空间下,生成式智能体的自动寻路寻路、环境以及其他角色检测等算法均只需考虑二个维度的向量坐标,构造方式较为简单,构造成本较低。但是二维空间与三维空间相比存在诸多劣势,比如:只提供平面图像而缺少深度信息、降低用户的空间感和沉浸感、限制智能体的行动模式、难以实现智能体的精细动作等。因此创造三维空间下的生成式智能体存在其必要性。
实现思路