BlocksWorld++

该数据集名为BlocksWorld++，它是对传统BlocksWorld问题的扩展，在模拟环境中涉及堆叠和拆解积木以匹配特定配置。该数据集还包括与感知和随机运动控制相关的挑战，并整合了AprilTag标记以进行物体识别。此外，它还提供了一个难度逐渐增加的任务课程，以评估持续学习的能力。这项任务的名称是“在真实硬件上使用BlocksWorld++课程进行持续学习评估”。

This dataset, named BlocksWorld++, is an extension of the classic BlocksWorld problem. It involves stacking and disassembling blocks in a simulated environment to match a target configuration. It also includes challenges related to perception and stochastic motion control, and integrates AprilTag markers for object recognition. Additionally, it provides a task curriculum with gradually increasing difficulty to evaluate continuous learning capabilities. The name of this evaluation task is "Continuous Learning Evaluation with BlocksWorld++ Curriculum on Real Hardware".

搜集汇总

背景与挑战

背景概述

BlocksWorld++是一个机器人任务规划的持续学习框架数据集，展示了CLIMB框架如何通过基础模型和执行反馈构建PDDL模型，并在简单和复杂任务中实现零-shot规划及自我改进。数据集还包括了在真实硬件上的演示，验证了其从模拟到现实的转换能力。