《Crowdsourcing for closed-loop control》由会员分享，可在线阅读，更多相关《Crowdsourcing for closed-loop control（4页珍藏版）》请在金锄头文库上搜索。
1、Crowdsourcing for closed-loop controlSarah OsentoskiDepartment of Computer ScienceBrown UniversityProvidence, RI 02912sosentoscs.brown.eduChristopher CrickDepartment of Computer ScienceBrown UniversityProvidence, RI 02912chriscrickcs.brown.eduGrayin JayDepartment of Computer ScienceBrown UniversityP
2、rovidence, RI 02912tjaycs.brown.eduOdest Chadwicke JenkinsDepartment of Computer ScienceBrown UniversityProvidence, RI 02912cjenkinscs.brown.eduAbstractWe present a system for large scale robotic learning from demonstration. We de-scribe a set of software tools for enabling human-robot interaction o
3、ver the inter-net and gathering the large datasets that such crowdsourcing makes possible. Weshow results in which humans teach a robot to navigate a maze over the Internet.Robots occupy a peculiar place in our culture. We have been building robots in our imaginationsfor decades, robots that are alt
4、ernately wondrous or terrifying, always brilliant and consummatelyskilled. In contrast, real robots are typically brittle and capable of only a few simple constrainedtasks. Additionally, only expert programmers, intimately familiar with the particulars of a low-levelrobotic system, can hope to achie
5、ve any kind of complex robot behavior. This paper describes ourefforts to apply the lessons of crowdsourcing to robotics to leverage the power and knowledge of atruly large number of end users to create more skilled and robust robot controllers.We focus on learning from demonstration (LfD), 2, 1 an
6、approach to robot programming in whichusers demonstrate desired skills to a robot. Nothing is required of the user beyond the ability tocomplete the task in a way that the robot can interpret. Traditionally, LfD research has been con-strained by the number of demonstrations that can be performed; un
7、less a large number of userscan interact with the robot only a limited amount of data can easily be gathered. Since users do notusually need specialized skills to demonstrate robot skills, a web-enabled system could be used tocollect data from a large number of users. An online system also lifts the
8、 burden of training therobot from a single user who may only want to contribute a few demonstrations. Collecting datafor closed-loop control is different from many crowdsourcing applications examined in the MachineLearning community, which has typically focused on annotation of text and labeling of
9、images. Taskdemonstration often requires a significant interaction both in terms of time and information providedto the user. Users not only give the robot instructions, but also evaluate the results and provide newinstructions given the outcome.We describe a recently developed system that allows a
10、large number of users to train a robot tosolve a task (in this case maze-navigation) through a video-game style interface. While a singledemonstration may contain errors and provides only limited data, the demonstrations from multipleusers provide enough data to create a robust policy. There have be
11、en a few initial efforts in puttingrobots on the Internet 4, 9. These approaches generally allowed people to interact with robots butwere not aimed at task learning. Other work examined using crowdsourcing approaches to trainrobots through game playing environments. Chernova et al examined using a m
12、ulti-player videogame where users collaborate to provide user demonstrations 3. This work differs from ours in that1it relied on real-world user interactions in a museum, rather than an online setting with a potentiallyglobal user base. Additionally, users were not providing demonstration on the act
13、ual robot. Oursystem allows users to actually control the robot and does not require special software or pluginsother than a web browser.1 Robot and Web InterfaceThe robot used in our experiments is an iRobot Create with a FitPC2 small-form-factor computerand a Sony PS3 Eye camera. The computer main
14、tains a wireless connection to the Internet allowinguser interactions. The robot is able to move forward, backwards, and rotate. The robot can maneuverthrough a maze, pictured in Figure 1a The maze has artificial reality (AR) tags placed within it aslandmarks. The robot is able to detect the AR tags
15、 and use the size of the tag in its visual perceptionto estimate its distance from each tag. The tags, along with the bump sensors that detect collisionswith walls, represent the perceptual space of the robot.The system builds upon ROS 7, Willow Garages robot middleware system and leverages rosjs6,
16、a lightweight Javascript binding for ROS. rosjs exposes the robots functionality and sensors asweb services, as well as providing security and visualization tools. Robot application developersand researchers can create robot controllers and interfaces in the same manner as creating webcontentrosjs does not require users to install additional software or plugins beyond a web browser,allowing a large number of users the ability to access the robot. Figure 1 shows the two inter
