Visibility-based spatial reasoning for object manipulation in cluttered environments
Introduction
Recently, a wide spectrum of effort has been expended in order to extend the robotics technology beyond industrial applications. Examples include mobile robots in the realm of service and personal assistance, especially for aiding the elderly or the disabled. A typical task of such a service or assistant robot is to manipulate objects of daily necessities on the user’s request. Note that object manipulation comprises several subtasks, such as approaching, grasping, removing and delivering, each of which appears to be highly involved. In general, it is extremely difficult for a robot to achieve full autonomy of object manipulation in a home or service environment. This is because, unlike an industrial environment, a home or service environment is often unstructured, may not be kept under control, and thus, may not remain static. Furthermore, object manipulation in the context of service to humans needs to be performed in real time.
Traditionally, object manipulation has been part of motion planning. Unfortunately, the complex nature of motion planning for high DOF (degrees of freedom) robots often makes it impossible to achieve real-time performance [1]. For instance, the generation of a mobile manipulation path for a humanoid robot to grasp an object may require a few minutes [2]. Even for a single arm (typically of 6 or 7 DOF) with no mobile platform, real-time motion planning has been a challenging task [3]. It becomes much harder, if not intractable, for motion planning to be done in highly cluttered environments. The direct application of conventional motion planning to a grasp point selected on a trial-and-error basis would be too costly and prone to failure if the selected grasp point of the requested object does not turn out to be collision-free. If the grasp point is not reachable due to the obstacles in the neighborhood of the object, the robot requires further time-consuming motion planning, with a new grasp point until it succeeds. It would be much better to have a local but fast motion planner that can simultaneously determine a grasp point and local paths to access the grasp point and remove the grasped object, such that the local motion planner can be well integrated with a general motion planner for improved overall efficiency.
This paper proposes real-time techniques for providing such a local motion plan for the cluttered neighborhood of a requested object, and leaving the rest to a general motion planner. Specifically, this paper focuses on the visibility-based spatial reasoning techniques, in order to determine the gripper access directions and object removal directions, i.e. the directions along which the robot gripper can access and remove the requested object. The real-time performance is achieved using visibility query and environment map, which are supported by commodity graphics hardware.
The organization of this paper is as follows. Section 2 introduces related work. Section 3 presents object recognition and 3D workspace modeling. Section 4 overviews the proposed approach in the context of motion planning. The major contributions of this paper are presented in Section 5, which discusses the gripper accessibility analysis, and Section 6, which discusses the technique for determining the object removal directions. Section 7 discusses the integration of the proposed approach with general motion planning. Section 8 presents the experimental results and evaluates the performance of the proposed approaches. Finally, Section 9 concludes this paper.
Section snippets
Related work
The visibility-based spatial reasoning techniques presented in this paper complement general motion planning, and eventually enable the motion planner to achieve real-time performance in cluttered environments. In the motion planning field, probabilistic/randomized methods have been targeted to generate paths with high probability for high-dimensional configuration spaces, but may not be able to generate paths in real time. For instance, Kuffner et al. [2] used RRT-connect (Rapidly-exploring
Object recognition and workspace modeling
The experimental robotic platform used in this study is shown in Fig. 1(a). It is equipped with a robotic manipulator (Fig. 1(b)), which is similar to a human arm with 7 DOF. The arm contains a parallel jaw gripper (Fig. 1(c)), which provides the basic capability for grasping and manipulating objects. A stereo camera is mounted on the parallel jaw gripper in an eye-on-hand configuration, and is used for modeling the workspace.
Fig. 2 shows the overall software system built for object
Overview: Spatial reasoning for motion planning
The approach proposed in this paper delimits the cluttered environment neighboring the target object, as shown in Fig. 5. The cluttered neighborhood is represented by a sphere, named the locality sphere. Using the locality sphere, the overall path of the robot gripper to access, grasp, and deliver the target object is broken down into 4 sub-paths, as illustrated in Fig. 5: (i) from the start pose to the access pose, located at the boundary of the locality sphere, (ii) from the access pose to
Gripper accessibility analysis
This section presents the technique for computing the access pose, the grasp pose, and the access direction. As discussed in Section 3, all target objects have object models in the database, and the object model contains gripper accessibility information of each object. For the pizza sauce bottle shown in Fig. 6, it is reasonable to define four accessible directions: and with respect to its local frame. (Among them, only two, and , are illustrated in the figure). In contrast, access
Object removability analysis
In computing the collision-free removal directions, environment map [28] and Minkowski sum [29] prove to be useful. In computer graphics, the environment map is used to describe the scene surrounding an object. The most popular implementation of the environment map is the cube map [17], shown in Fig. 11, where each face covers a field of view both horizontally and vertically. There are six faces per cube, and each face is implemented as a square 2D image texture. Notice that a pixel (more
Discussions
In the current study, the elastic strip method is used for general motion planning, and is accelerated by gripper accessibility analysis (discussed in Section 5) and object removability analysis (discussed in Section 6) within the locality sphere, as illustrated in Fig. 5. The accessibility analysis is used to determine the set of all collision-free linear paths of the gripper from the access poses to the grasp poses. Without accessibility analysis, the elastic strip-based motion planner will
Experiments
The experimental mobile manipulator consists of a 7-DOF Amtec lightweight arm (Amtec Robotics GmbH) with a parallel jaw gripper installed on top of an Active Media Robotic’s Powerbot. The algorithms presented in this paper have been implemented on a modest PC (Pentium 4 2.8 GHz, and NVIDIA GeForce 6600GT graphics card). Fig. 14 shows a series of snapshots for the manipulation task, in which the robot is requested to grasp a pizza sauce bottle in Robot Café. Fig. 15 shows the corresponding
Conclusion
This paper presents a novel approach to accessibility and removability analyses for object manipulation tasks: visibility-based spatial reasoning. The accessibility and removability analyses utilize the visibility query and cube map, which are accelerated by graphics hardware. The performance and robustness of the proposed approach are evaluated in cluttered indoor environments. The experimental results demonstrated that the proposed methods are sufficiently fast and robust to manipulate 3D
Acknowledgements
The preliminary versions of the sections in this paper have appeared in [33], [34], [35]. The work presented in this paper was performed for the 21st Century Frontier R&D Programs funded by the Ministry of Science and Technology of Korea. This work was also supported by grant No. R01-2006-000-11297-0 from the Basic Research Program of the Korea Science & Engineering Foundation and by MIC, Korea under ITRC IITA-2006-(C1090-0603-0046).
References (35)
Geometric reasoning about assembly tools
Artificial Intelligence
(1998)- et al.
Efficient geometric algorithms for workpiece orientation in 4- and 5-axis NC machining
Computer-Aided Design
(1996) - Reif H. Complexity of the Mover’s problem and generalizations. In: Proc. of the 20th IEEE symposium on foundations of...
- Kuffner J, Nishiwaki K, Kagami S, Inaba M, Inoue H. Motion planning for humanoid robots. In: Proc. of the 11th...
- Hsu D, Latombe J, Kurniawati H. On the probabilistic foundations of probabilistic roadmap planning. In: Proc. of the...
- et al.
On finding narrow passages with probabilistic roadmap planners
- Hsu D, Jiang T, Reif J, Sun Z. The bridge test for sampling narrow passages with probabilistic roadmap planners. In:...
- Boor V, Overmars M, Stappen A. The gaussian sampling strategy for probabilistic roadmap planners. In: Proc. of the IEEE...
- Brock O. Generating robot motion: the integration of planning and execution. Ph.D. thesis. Stanford University;...
- Hong S, Moradi H, Lee S, Lee S. A comparative study on local minima escape in potential field method. In: Proc. of the...