LarsJohaNV/robot-fixture-insertion-cosmos-smoke-20260427

机器人固定件插入Cosmos成功数据集是一个二分类（是/否）视觉成功分类数据集，专为Cosmos-Reason微调设计。任务是通过单个近距离RGB图像判断机器人是否成功将黑色塑料固定件插入匹配的槽中。成功的定义是黑色塑料件以预期的最终姿势正确安装在匹配的槽中；失败的插入包括未安装、错位、部分或明显不成功的插入尝试。数据集包含训练、验证和测试分割。

The robot_fixture_insertion Cosmos Success Dataset is a binary yes/no visual success-classification dataset for Cosmos-Reason fine-tuning. The task is to classify whether the robot successfully inserted the black plastic fixture piece into the matching slot from a single close-up RGB image. Success is defined as the black plastic piece being seated in the matching slot in the intended final pose; failed insertions include unseated, misaligned, partial, or visibly unsuccessful insertion attempts. The dataset includes splits for training, validation, and testing.