novastar112/sokoban_easy_1box_remap_prev

VisGym sokoban_easy_1box Remap-Prev 数据集是一个由VisGym Sokoban环境生成的求解器成功轨迹集合。该数据集针对Sokoban游戏中的简单级别设计，环境参数包括：动作模式为仅移动（move_only），房间维度为[7, 7]，最大步数为20，箱子数量为1，生成步数为23，成功模式为stop_then_solved。动作模式包括移动动作（1到4）和停止动作（stop）。成功语义要求先停止再解决。图像策略中，每个历史步骤的image_prev等于当前图像，最终步骤的image_next等于当前图像。该数据集旨在用于Sokoban CE+MSE实验的remap-prev训练设置。VisGym sokoban_easy_1box Remap-Prev dataset consists of solver-success Sokoban trajectories generated from the VisGym Sokoban environment. It is designed for the task sokoban_easy_1box with environment ID sokoban/easy. Environment parameters include: action mode as move_only, room dimensions [7, 7], max steps 20, number of boxes 1, number of generation steps 23, and success mode stop_then_solved. The action schema is move-only with actions (move, 1..4) plus (stop, stop). Success semantics follow stop_then_solved. The image policy specifies that every history step has image_prev == image, and the final step has image_next == image. The dataset is intended for the remap-prev training setup used in Sokoban CE+MSE experiments.