According to data from the Aiqicha App, Alipay (Hangzhou) Digital Service Technology Co., Ltd. has recently disclosed an invention patent named "Control Method, Model Training Method, Device, and Storage Medium for Embodied Agents." The patent details a process wherein, when an embodied agent undertakes a target task, it gathers observational environmental data at the present moment. This data, combined with the operational data of the embodied agent from various historical time points during the target task, is then fed into a pre-trained autoregressive prediction model. Leveraging this model, the system can forecast environmental data for the subsequent time step. Based on this forecast, the system deduces the optimal action for the embodied agent to perform, thereby guiding it to successfully complete the task.
