The International Journal of Robotics Study, Ahead of Publish.
Support discovering (RL), replica discovering (IL), and job and movement preparation (TAMP) have actually shown excellent efficiency throughout different robot adjustment jobs. Nevertheless, these strategies have actually been restricted to finding out basic actions in present real-world adjustment standards, such as pressing or pick-and-place. To allow even more facility, long-horizon actions of an independent robotic, we recommend to concentrate on real-world furnishings setting up, a facility, long-horizon robot adjustment job that needs attending to numerous present robot adjustment obstacles. We offer FurnitureBench, a reproducible real-world furnishings setting up standard targeted at supplying a reduced obstacle for entrance and being conveniently reproducible, to make sure that scientists throughout the globe can accurately evaluate their formulas and contrast them versus previous job. For simplicity of usage, we give 200+ hours of pre-collected information (5000+ presentations), 3D furnishings designs, a robot atmosphere configuration overview, and organized job initialization. Moreover, we give FurnitureSim, a quick and sensible simulator of FurnitureBench. We benchmark the efficiency of offline RL, IL, and offline-to-online RL formulas on our setting up jobs and show the requirement to boost such formulas to be able to resolve our jobs in the real life, supplying adequate chances for future research study.
发布者：Minho Heo，转转请注明出处：https://robotalks.cn/furniturebench-reproducible-real-world-benchmark-for-long-horizon-complex-manipulation/
