S&AS: INT: COLLAB: Composable and Verifiable Design for Autonomous Humanoid Robots in Space Missions

S

• 批准号: 1724257
• 负责人: Robert Platt
• 金额: $ 46万
• 依托单位: Northeastern University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2017
• 资助国家: 美国
• 起止时间: 2017-09-01 至 2021-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1724257&HistoricalAwards=false
• 关键词: S& INT; COLLAB; Composable; Verifiable

Future space missions will increasingly rely on autonomous robots like the NASA Valkyrie human-centered robot for deploying equipment, assisting astronauts, and maintaining facilities in real world partially-observable and cluttered environments. Despite significant progress in robotic mobility, manipulation, and perception, there has been relatively little progress on providing formal performance guarantees for these integrated systems. Formal guarantees are critical for achieving long term autonomy, particularly for robots performing complex tasks requiring successful execution of multiple component subtasks. Thus, the goal of this project is to develop performance guarantees for space robots operating in unstructured real world environments. Although robots are used as design examples, the project is of a basic research nature and the results can have impacts on other fields, such as sensor/actuator networks, manufacturing and transportation systems. The multidisciplinary approach taken for this project will help broaden participation of underrepresented groups and positively impact engineering and computer science education.The objective of this project is to develop new methods to synthesize coordinated manipulation and locomotion plans and control policies that verifiably adhere to formal mission specifications. There are two major thrusts. First, the PIs plan to develop manipulation, locomotion, and motion primitives that can provide performance guarantees in unstructured, partially observable, and dynamic environments. The focus will be on using methods from perception and planning under uncertainty to provide guarantees in cluttered and partially observable environments. The PIs will also leverage new tools from hybrid systems and sampling based methods to achieve controllers with verifiable guarantees through contact mode switches. Second, the PIs plan to devise methods to automatically synthesize mission plans in a way that can guarantee the accomplishment of high-level mission goals or bound the probability of failure. The focus will be on automatic and learning-based design, enabling the system to adapt to changing environments, uncertain faults and potential adversaries. Most of the work performed under this project will be demonstrated in the context of complex space tasks inspired by NASA scenarios.

未来的太空任务将越来越多地依赖自主机器人，比如NASA的Valkyrie以人为中心的机器人，在现实世界中部分可观察和混乱的环境中部署设备、协助宇航员和维护设施。尽管在机器人移动性、操纵和感知方面取得了重大进展，但在为这些集成系统提供正式性能保证方面进展相对较少。正式保证对于实现长期自治至关重要，特别是对于执行需要成功执行多个组件子任务的复杂任务的机器人。因此，该项目的目标是为在非结构化的真实世界环境中运行的空间机器人开发性能保证。虽然机器人被用作设计示例，但该项目具有基础研究性质，其结果可能对其他领域产生影响，例如传感器/执行器网络，制造和运输系统。该项目的多学科方法将有助于扩大代表性不足的群体的参与，并对工程和计算机科学教育产生积极影响。该项目的目标是开发新的方法来综合协调的操纵和运动计划以及控制策略，可验证地遵守正式的任务规范。有两个主要的推动力。首先，pi计划开发操作、运动和运动原语，这些原语可以在非结构化、部分可观察和动态环境中提供性能保证。重点将是在不确定性下使用感知和规划方法，在混乱和部分可观察的环境中提供保障。pi还将利用混合系统的新工具和基于采样的方法，通过接触模式开关实现具有可验证保证的控制器。二是计划设计任务计划自动合成的方法，以保证高级别任务目标的完成或限制失败的概率。重点将放在自动和基于学习的设计上，使系统能够适应不断变化的环境、不确定的故障和潜在的对手。在该项目下开展的大部分工作将在受NASA情景启发的复杂空间任务背景下进行演示。

项目成果

Learning Multi-Level Hierarchies with Hindsight

• 发表时间: 2017-12
• 作者: Andrew Levy;G. Konidaris;Robert W. Platt;Kate Saenko

SO(2)-Equivariant Reinforcement Learning

• DOI: 10.48550/arxiv.2203.04439
• 发表时间: 2022-03
• 期刊: ArXiv
• 作者: Dian Wang;R. Walters;Robert W. Platt

Policy learning in SE (3) action spaces

SE (3) 行动空间中的政策学习

• 发表时间: 2020
• 期刊: Proceedings of the Conference on Robot Learning
• 作者: Wang, Dian;Kohler, Colin;Platt, Robert
• 通讯作者: Platt, Robert

Grasp Pose Detection in Point Clouds

• DOI: 10.1177/0278364917735594
• 发表时间: 2017-12-01
• 期刊: INTERNATIONAL JOURNAL OF ROBOTICS RESEARCH
• 影响因子: 9.2
• 作者: ten Pas, Andreas;Gualtieri, Marcus;Platt, Robert
• 通讯作者: Platt, Robert

Learning Bayes Filter Models for Tactile Localization

学习用于触觉定位的贝叶斯滤波器模型

• 发表时间: 2020
• 期刊: Proceedings of the IEEERSJ International Conference on Intelligent Robots and Systems
• 作者: Kelestemur, Tarik;Keil, Colin;Whitney, John;Platt, Robert;Padir, Taskin
• 通讯作者: Padir, Taskin