在用于物体操作的生成式世界模型中表示位置信息

Stefano Ferraro, Pietro Mazzaglia, Tim Verbelen, Bart Dhoedt, Sai Rajeswar
{"title":"在用于物体操作的生成式世界模型中表示位置信息","authors":"Stefano Ferraro, Pietro Mazzaglia, Tim Verbelen, Bart Dhoedt, Sai Rajeswar","doi":"arxiv-2409.12005","DOIUrl":null,"url":null,"abstract":"Object manipulation capabilities are essential skills that set apart embodied\nagents engaging with the world, especially in the realm of robotics. The\nability to predict outcomes of interactions with objects is paramount in this\nsetting. While model-based control methods have started to be employed for\ntackling manipulation tasks, they have faced challenges in accurately\nmanipulating objects. As we analyze the causes of this limitation, we identify\nthe cause of underperformance in the way current world models represent crucial\npositional information, especially about the target's goal specification for\nobject positioning tasks. We introduce a general approach that empowers world\nmodel-based agents to effectively solve object-positioning tasks. We propose\ntwo declinations of this approach for generative world models:\nposition-conditioned (PCP) and latent-conditioned (LCP) policy learning. In\nparticular, LCP employs object-centric latent representations that explicitly\ncapture object positional information for goal specification. This naturally\nleads to the emergence of multimodal capabilities, enabling the specification\nof goals through spatial coordinates or a visual goal. Our methods are\nrigorously evaluated across several manipulation environments, showing\nfavorable performance compared to current model-based control approaches.","PeriodicalId":501031,"journal":{"name":"arXiv - CS - Robotics","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Representing Positional Information in Generative World Models for Object Manipulation\",\"authors\":\"Stefano Ferraro, Pietro Mazzaglia, Tim Verbelen, Bart Dhoedt, Sai Rajeswar\",\"doi\":\"arxiv-2409.12005\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Object manipulation capabilities are essential skills that set apart embodied\\nagents engaging with the world, especially in the realm of robotics. The\\nability to predict outcomes of interactions with objects is paramount in this\\nsetting. While model-based control methods have started to be employed for\\ntackling manipulation tasks, they have faced challenges in accurately\\nmanipulating objects. As we analyze the causes of this limitation, we identify\\nthe cause of underperformance in the way current world models represent crucial\\npositional information, especially about the target's goal specification for\\nobject positioning tasks. We introduce a general approach that empowers world\\nmodel-based agents to effectively solve object-positioning tasks. We propose\\ntwo declinations of this approach for generative world models:\\nposition-conditioned (PCP) and latent-conditioned (LCP) policy learning. In\\nparticular, LCP employs object-centric latent representations that explicitly\\ncapture object positional information for goal specification. This naturally\\nleads to the emergence of multimodal capabilities, enabling the specification\\nof goals through spatial coordinates or a visual goal. Our methods are\\nrigorously evaluated across several manipulation environments, showing\\nfavorable performance compared to current model-based control approaches.\",\"PeriodicalId\":501031,\"journal\":{\"name\":\"arXiv - CS - Robotics\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-18\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - CS - Robotics\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.12005\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Robotics","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.12005","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

摘要

物体操作能力是使与世界打交道的化身机器人与众不同的基本技能,尤其是在机器人领域。在这种情况下,预测与物体交互结果的能力至关重要。虽然基于模型的控制方法已开始用于处理操纵任务,但它们在准确操纵物体方面面临挑战。在分析造成这种限制的原因时,我们发现性能不佳的原因在于当前世界模型表示关键位置信息的方式,特别是关于目标的目标定位任务。我们介绍了一种通用方法,它能让基于世界模型的代理有效地解决物体定位任务。我们为生成式世界模型提出了两种方法:位置条件(PCP)和潜伏条件(LCP)策略学习。其中,LCP 采用以物体为中心的潜在表征,明确捕捉物体位置信息,用于目标指定。这自然会导致多模态能力的出现,从而能够通过空间坐标或视觉目标来指定目标。我们的方法在多个操纵环境中进行了理论评估,显示出与当前基于模型的控制方法相比更优越的性能。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
Representing Positional Information in Generative World Models for Object Manipulation
Object manipulation capabilities are essential skills that set apart embodied agents engaging with the world, especially in the realm of robotics. The ability to predict outcomes of interactions with objects is paramount in this setting. While model-based control methods have started to be employed for tackling manipulation tasks, they have faced challenges in accurately manipulating objects. As we analyze the causes of this limitation, we identify the cause of underperformance in the way current world models represent crucial positional information, especially about the target's goal specification for object positioning tasks. We introduce a general approach that empowers world model-based agents to effectively solve object-positioning tasks. We propose two declinations of this approach for generative world models: position-conditioned (PCP) and latent-conditioned (LCP) policy learning. In particular, LCP employs object-centric latent representations that explicitly capture object positional information for goal specification. This naturally leads to the emergence of multimodal capabilities, enabling the specification of goals through spatial coordinates or a visual goal. Our methods are rigorously evaluated across several manipulation environments, showing favorable performance compared to current model-based control approaches.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信