Flexible Control of Complex Kinematic Chains - Semantic Scholar

5 downloads 1117 Views 2MB Size Report
aspects of control architectures and, most importantly, of the programming ...... during several years, some of which coming from DLR and eventually hiring at.
Flexible Control of Complex Kinematic Chains Rainer Bischoff, Günter Schreiber, Bernd Finkemeyer, Yevgen Kogan, Marinus Danzer, and Johannes Kurth*

Abstract. Robots providing meaningful services in unconstraint environments require advanced mobility, manipulation, sensing and perception capabilities. KUKA’s task in DESIRE was to develop robot arms and control hardware and software that are flexible with respect to the number of degrees of freedom and their connections to build kinematic chains and the type of sensors used to perceive the environment. To enable the DESIRE project partners and research partners beyond the DESIRE consortium to work with advanced lightweight torquecontrolled robot arms KUKA and DLR engaged in a technology transfer leading to the commercialization of the DLR Lightweight Robot. After many innovative steps, first at DLR beginning in the early 1990s, later at DLR and KUKA, both partners managed to successfully go the strenuous road from the original invention, an idea made manifest in 1991, to prototypes produced in a small series starting in December 2008 [5]. This paper first reviews the most important steps leading to this development. Second, the development of flexible controllers which enabled the project partners to integrate the KUKA-DLR Lightweight Robot in a dual-arm robot system is presented. Furthermore, independent KUKA demonstrators – an omnidirectional mobile manipulator and a dual arm system – are introduced which demonstrate the capabilities of a newly developed controller framework.

1 Introduction The main motivating force behind the KUKA developments in the cooperative research project DESIRE is to revolutionize the applicability of robotics in our society. Robots should become available not only on the shop floor, but also at our homes and offices, and in the public and in space. They have to collaborate with Rainer Bischoff · Günter Schreiber · Bernd Finkemeyer · Yevgen Kogan · Marinus Danzer · Johannes Kurth KUKA Roboter GmbH, Augsburg, Germany e-mail: {Rainer.Bischoff,Günter Schreiber,Bernd Finkemeyer, Yevgen Kogan,Marinus Danzer,Johannes Kurth}@kuka.com E. Prassler et al. (Eds.): Towards Service Robots for Everyday Environ., STAR 76, pp. 331–352. © Springer-Verlag Berlin Heidelberg 2012 springerlink.com

332

R. Bischoff et al.

humans, cooperate among themselves, work safely, intelligently and largely autonomously, and all this in environments, which are not any longer pre-defined, but probably only loosely structured or not structured at all. Developing robots and controllers for both for industrial and service application is strongly connected to the new capabilities to be displayed within unstructured environments where sensing and compliant motion control become important aspects of control architectures and, most importantly, of the programming methodology. Looking at the future of automation, robots will not only be simple and senseless machines carrying out dirty, dull and dangerous work and being caged behind fences, but work as skillful robot assistants in close proximity of, and in cooperation with, humans. These robot assistants will require the following characteristics in sharp contrast to today’s industrial robots: frequent interaction with humans and task changes, work space sharing, force/precision assistance, flexible relocation/mobile use, profitable even with small lotsizes and programmable by nonrobotics experts [2]. The same holds true for robot assistants at homes and offices, with the difference that non-robotics experts will rather instruct the robots to perform certain tasks than programming them, which means that “lotsize one” will become a commodity rather than an exception. The KUKA-DLR Lightweight Robot (LWR) is on the pathway to become a key element of future robot assistants at work and home environments because it is – due to its innovative characteristics – in principle less dangerous and easier programmable for tasks which require close human-robot interaction and is much more portable and energy-saving than robots with comparable payloads, and thus suitable, for mobile robot applications. In the context of the DESIRE project the first goal of the development was defined from a system perspective. It was the aim to push the development of the LWR towards an embeddable component. Towards this end its controllers (main controller and joint-level controllers) had to be improved from both a hardware and software point of view (see section 2). The second goal was formulated from a control perspective. In this respect the complexity of a service robot system is not only defined by the number of its actuators and sensors, but also by the heterogeneity of its actuators and sensors. The actual challenge was to develop a controller for these different actuators and sensors in such a way that it provides a framework for robust mobile manipulation (see section 3). The applicability of the developments was validated in two ways. First in the context of the common project demonstrator, two LWRs were mounted onto the DESIRE technology platform and intensively used for manipulation tasks with three- and four-finger hands while avoiding collisions in real time (section 4.1). A second validation took place through independent KUKA demonstrations. Section 4.2 reports on the haptic guidance of a mobile manipulator, the KUKA omniRob, which consists of an omnidirectional mobile base and an LWR with a total number of 10 degrees of freedom. In this demonstration the LWR was used as an advanced input device for steering the mobile platform. Based on the same controller framework a separate demonstrator, a dual LWR system, was set up and controlled (section 4.3).

Flexible Control of Complex Kinematic Chains

333

2 KUKA-DLR Lightweight Robot 2.1 From Research to Technology Transfer and Product Development Within a series of collaborative funded research projects and KUKA independent developments we contributed to the development of the KUKA lightweight robot (LWR) arm as stand-alone product and, in particular, as a component that can be easily integrated into other robotic systems. Inspired by the HERMES project [3] integration was brought forward on the three required levels: mechanical, electrical and data (interface) integration. The PAPAS project, sponsored by the German Federal Ministry of Education and Research (BMBF), was initiated in 2003 and was concerned with the development of plug-and-play drive and control technologies (see [27]), a mechatronic development environment (see [16]) and with combining the advanced control of the LWR developed by DLR with an industry-proven controller that could provide a programming and operation environment and sequence control developed by KUKA [2]. During the project, the DLR basic controller and the KUKA controller were for the first time connected with each other via Ethernet. The KUKA teach pendant provided the already established programming and operation environment and look & feel for the industrial user and at the same time enabled the access to the new lightweight robot technology with its unique performance characteristics. The DLR controller was used as the external clock for the KUKA controller which could adjust to this external timing. Both controllers communicated asynchronously and synchronously at the interpolation rate of the KUKA controller. Asynchronous commands were used to convey impedance and redundancy parameters, while interpolated joint angles and the Cartesian position of the flange were synchronously communicated. Thus, the KUKA controller could provide sequence control, industrially relevant I/O communication through field busses and offer the user an industry-proven interface for programming and operation, and for the first time, the DLR compliance control could be made available for application programming (Figure 1). On the occasion of “Automatica 2004” (the world’s largest robot trade fair) the “RoboAssistant” – this is how the combination of DLR LWR and KUKA controller was named – was presented to the expert visitors for the first time. Programming by demonstration and joining various bolts were successfully demonstrated. The visitors were allowed to manually move and program the robot as described in [11]. The vision of a robot assisting a worker during production processes was thus becoming obvious for the visitors. Some visitors rated the different characteristics of the lightweight robot in a questionnaire. Programming by demonstration, compliance, and the robot’s internal cable routing were named as the three most outstanding features [4].

334

R. Bischoff et al.

Fig. 1 Coupling of KUKA’s PC-based standard robot controller (KRC) and operator interface (KUKA Control Panel KCP) with DLR’s PC-based controller for the Lightweight Robot LWR.

The results of the PAPAS project and the public presentation were very encouraging for KUKA and DLR and led to increased efforts in the technology transfer. Once again the BMBF supported this technology transfer through the project DESIRE [8] which had the overall objective of maintaining and extending the leading role of German industry and research in the field of service robotics. One major aim of the project was to increase the everyday applicability of perception and manipulation and to integrate the majority of the components developed in the project to a common technology platform. One part of the project was designed as a PAPAS follow-up technology transfer project between DLR and KUKA. The two-PC controller solution originating from PAPAS was developed further into a one-PC controller solution (based on the KUKA Robot Controller KRC). Here, a communication data buffer that caused dead times in the two-PC solution could be eliminated. To increase the performance of the whole system the robot’s impedance control was further improved. The first DESIRE technology platform was equipped early 2007 with two LWR (third generation) and the single KRC controller solution. The project partners were trained on the special features and the new KUKA Robot Language (KRL) commands to program the robot in February 2007. Based on these developments and experiences gathered with the project partners the fourth generation LWR was developed outside the project. A production of a mini-series of 60 LWR4 started in December 2008. Within 12 months all 60 robots were either sold or lend to research partners. For the final DESIRE demonstrations two LWR4 were made available to the project partners and integrated into the DESIRE technology platform; another three LWRs were used in KUKA’s additional DESIRE

Flexible Control of Complex Kinematic Chains

335

demonstrators (see section 4). The currently available LWR version 4+ continues to be successful on the market, and is even used in production lines [7].

2.2 Innovative Characteristics of the KUKA Lightweight Robot The characteristics of the LWR3 (Figure 2) and LWR4 are comparable and are based on concepts which are generally regarded as decisive for the next generation of robots that are to be capable of working together with humans. The weight was reduced to the limits of what is technically possible, which decisively improves the robot’s dynamic performance. This makes it also safer than other robots and increases its applicability in complex production processes, such as assembly. The lightweight robot LWR4 is designed for a rated payload of 7 kg, and itself has a mass of 15 kg. Its low mass helps reduce the power consumption and additionally allows a hitherto unknown degree of mobility for robot arms. In the first place, the robot can be carried manually to its place of use, and secondly, battery-powered operation is possible in mobile robot systems, for example. With its seven axes, the robot has one redundant degree of freedom, which gives the programmer more flexibility in cluttered workspaces. The seven axes also help to avoid typical singularities of 6-axis kinematic systems. The rounded design, which rules out any risk of crushing between structural components, contributes to the overall safety [12].

Fig. 2 The KUKA Lightweight Robot LWR3 with controller (2006). The only visible difference of the LWR3 to the LWR4 is the modified cable routing at the foot of the robot, which is pluggable with the LWR4.

336

R. Bischoff et al.

Torque sensors in each of the seven joints, a detailed dynamic model of the robot, state control and a high servo-control cycle rate (3 kHz locally in the joints, 1 kHz overall), combined with powerful drives and the lightweight construction, enable active damping of vibrations to achieve excellent motion performance (path accuracy, repeatability) [1]. Furthermore, this also makes it possible to achieve a programmable compliance, both axis-specific and Cartesian [19]. This allows the robot to act like a spring-damper system in which the parameters can be set within wide limits. This compliance control enables the robot to be manually guided, thereby opening up a totally new experience in human-robot interaction. A programmer or user can thus move the robot intuitively and quickly to the desired position. A further advantage is to enable the programming of assembly procedures that could previously be implemented only with great difficulty. Moreover, it is no longer necessary to use compliant grippers or other equipment, as the arm already provides the required compliance. Control parameters can be switched over within one control cycle (1 ms). In this way, it is possible to switch extremely quickly from a stiff, position-controlled mode to a compliant behavior. The high sensitivity of the lightweight arm and the detailed knowledge of the model allow detection of collisions. This sensitivity, coupled with the advanced servo control, enables faster joining of work pieces since it is possible to move on the programmed path right up to a planned collision with the component, and then search for an edge or hole in compliant mode. In this way, the time taken to execute an assembly task can be significantly reduced.

3 Flexible Control for Service Robots 3.1 State of the Art Although industrial robots are multi-purpose and can be used in a huge variety of applications their controllers often only support simple trajectory following based on position control. Sometimes dedicated sensor interfaces exist that allow adapting the trajectory online. When analyzing service robots that are only capable of delivering one special service (e.g. vacuum cleaning) one must state that their robot controllers are much simpler than today’s industrial robot controllers – they are often just simple microcontrollers programmed for a dedicated task. Future robot assistants and personal robots, however, that will be capable of navigation and manipulation and that are able to cope with uncertainties of the environment through sensors will need much more sophisticated controllers. Currently, the robotics community is witnessing four trends with respect to robot controller hardware and software. First, industrial robot controllers are becoming more flexible by offering dedicated interfaces for (multi-)sensor integration. All major European brands such as ABB, Comau, and KUKA, but in the meantime also Japanese manufacturers Fanuc, Kawasaki, Mitsubishi and Motoman, are offering such interfaces. The openness, bandwidth and communication rate of these interfaces vary significantly, from path adaptations in interpolation cycle

Flexible Control of Complex Kinematic Chains

337

intervals to modifying controller strategy and parameters, sometimes even through an Application Programmer’s Interface [14]. Second, open source robot controllers which are very popular in pure research environments begin to leave their habitual environments in numbers. Good examples are OROCOS [18] and ROS [22], which are also used by industrial players for their predevelopments. Third, peripheral devices are not any longer connected by serial connections or slow real-time field busses. With advent of industrial Ethernet the field of automation is quickly changing. A new generation of high-performance servo-drives and sensors can be flexibly connected through real-time Ethernet-based field busses such as EtherCAT, POWERLINK, ProfiNet, and Sercos III. Standards that allow the plug-and-play of components through these busses are still missing, but research points in the right direction [27]. Fourth, programming of robot controllers is a tedious task. Expert knowledge is usually required, independent whether the robot is programmed online (by teaching points and trajectories) or offline (through CAD systems). Industrial robot makers typically offer training courses to teach robot users how to move their robots through space or in simulation and how to develop applications. This is needed because every robot maker offers its own programming or scripting language and different human-robot interaction devices. High-level programming languages, which could in principle be understood by today’s developers and which have large support through professional Integrated Development Environments (IDEs), only exist in research labs. Especially, when it comes to programming sensor-guided motions expert know-how in several programming domains is needed. Harmonized frameworks for programming sensors and robots and their interactions do not yet exist. In [9] a generalized controller/programming interface to execute complex sensor guided and guarded motions is presented. It serves as a good example of how such an interface should be designed on the basis of a robot assembly application. In conclusion, commercial robot controllers currently offer a great variety of operating systems, controller philosophies, programming languages and interfaces. It is unlikely that this variety will decrease in the near future. In contrast, as robotics is the art/science of integration, it will become extremely important for future robot controller architectures to offer the system integrator the possibility to quickly engineer and integrate software and hardware components that offer best performance or represent best practice for a given task. Programming new applications with complex kinematic chains and multiple sensory feedback needs to become much simpler than today. Therefore, it is important to clarify the principle structure of a robot controller which can cope with sensor feedback. In the following, a generalized robot controller is presented which consists of several modular hardware and software components and interfaces in-between (section 3.2). It is the basis for an approach to a new flexible KUKA controller framework (section 3.4). As an intermediate step towards this new framework several interfaces to the existing KUKA Robot Controller are presented that have been developed in recent research projects to fulfill the demands of researchers (section 3.3).

338

R. Bischoff et al.

3.2 Generalized Robot Controller KUKA’s center of gravity in the DESIRE project was the development of a flexible controller framework capable of controlling mobile redundant service robot kinematics in a variety of applications and domains, ranging from households to medical applications and industry. A prerequisite for such diversity is that the controller framework can cope with different kinematics and sensory equipment which is required for the task. It is likely that the required components of such a highly complex mechatronic system will not come from a single manufacturer, but from different ones, which means that hardware and software components of different manufacturers have to be integrated. Therefore, it is important to have a unifying controller basis which enables the integration of different drives, sensors, controllers and field busses. To advance the state of the art from position controlled robots this new controller framework aimed at the integration of different algorithms for kinematics, dynamics, redundancy and collision avoidance computations, compliance and hybrid control, and sensory feedbacks. As a basis for the discussion with the project partners the architecture of a generalized modular controller was developed early in the project according to Figure 3. It can be used to describe the control processes and hardware and software requirements in the context of a complex robotics research platform such as the DESIRE technology platform. A modular approach foresees interfaces on all levels: trajectory planning, motion planning/trajectory interpolation, motion execution and drive level. These interfaces are encapsulated by the “Manipulation API” which represents an application programmer’s interface to enable access to the sub-interfaces. Starting from the top in Figure 3 the user of this Manipulation API is able to set joint and Cartesian level set points as well as the interpolation type (such as PTP, LIN, CIRC and parameters such as maximum acceleration amax and velocity vmax) between these set points. Based on these set points in space a trajectory is planned which defines the geometry of a path in space. Adding a velocity profile to this geometric path and distributing the motion to the available joints will lead to an interpolated trajectory for the joints. A “drive manager” is then responsible to hand over the interpolated motions to the various drives of the robot. For example, a KUKA industrial robot will receive the joint set points through a dual-ported RAM (DP RAM), from where it is taken by a so-called digital servo electronics (DSE) module and fine interpolated before handed over to the drives. Similar transformations from joint set points to real drive set points would be computed by SCHUNK and Neobotix hardware and software drivers. The KUKA Lightweight Robot is connected through the Sercos bus and performs the fine interpolation in its joint electronics. Next to the drive control and drive manager there is also an I/O control and I/O manager which is responsible to distribute sensing tasks to the various devices through field busses, such as CAN bus, Profibus or Ethernet. Sensor information is fed back to the trajectory interpolation, e.g., by adapting a pre-planned path in such a way that a robot’s tool center point is following a weld seam.

Flexible Control of Complex Kinematic Chains

339

Trajectory interpolation and sensor feedback require real-time controllers and synchronous interfaces, which means that any output and input signal have to adapt to the clock of the real-time system, which is usually clocked at 1-12 millisecond intervals. Handing over set points through a trajectory planning interface or fully planned (geometric) trajectories through a motion planning interface does not require real-time interfaces, so that asynchronous access in the range of 200 ms is sufficient. Manipulation API trajectory planning interface asynchronous (>200 ms) trajectory planning level

motion planning interface

set points (Cartesian, joints); interpolation type (PTP, LIN, …)

trajectory planning geometric path

asynchronous (>200 ms)

motion planning level

trajectory interpolation synchronous (~ 1-12 ms)

Cartesian interface

drive control

joint set points in IPO cycle

joint interface synchronous (~ 1-12 ms) motion execution level

I/O control

interpolation inverse kinematics redundancy resolution Cartesian impedance control fast sensor feedback hybrid control

sensor data in sensor cycle

I/O manager

drive manager driver DP RAM

driver SCHUNK

driver CAN

DSE

PCI / DSP / FPGA

Neobotix HW

robot arm drives

finger drives

wheel drives

driver Sercos

driver field busses

driver SCHUNK

driver Ethernet

field bus

CAN / Profibus

Ethernet

sensor 1

hand

camera

bus interface synchronous (