FeelTheForce (FTF)

该数据集包含了针对五种力敏感操作任务的人类演示和远程操作演示，这些数据是为了训练一个能够模拟人类触觉行为的机器人学习系统，以实现精确控制。该数据集用于评估和比较FTF机器人学习系统与各种基准方法的性能。每个任务包含了30个演示（包括人类和远程操作）。这些任务包括将软面包放在盘子上、拆开塑料杯堆、将鸡蛋放入锅中、将薯片袋放在盘子上以及扭转并提起瓶盖等力敏感操作任务。This dataset contains human and teleoperation demonstrations for five force-sensitive manipulation tasks, collected to train a robotic learning system that mimics human tactile behaviors to achieve precise control. It is used to evaluate and compare the performance of the FTF robotic learning system against various baseline methods. Each task includes 30 demonstrations, covering both human and teleoperation scenarios. The specific tasks are as follows: placing soft bread onto a plate, disassembling a stack of plastic cups, putting eggs into a pot, placing chip bags onto a plate, and twisting and lifting bottle caps, all of which fall under the category of force-sensitive manipulation tasks.

搜集汇总

背景与挑战

背景概述

FeelTheForce (FTF) 是一个新颖的学习框架，专注于通过人类触觉演示实现机器人的力敏感操作学习。它使用低成本触觉手套和视觉系统捕获人类接触力数据，并训练闭环策略预测机器人轨迹和力，实现零样本从人类到机器人的转移，无需训练中的机器人数据，在多个操作任务中展示出高成功率和鲁棒性。