Method for controlling a robot device, robot device and computer program product

Abstract

A method for controlling a robot device (500) having a movable manipulator and/or effector (400), according to which method a speed and/or direction of movement of the manipulator and/or effector (400) is monitored and adjusted as appropriate, taking into consideration medical parameters for injury and robot dynamics is provided. A robot device (500) for implementing such a method and to a computer program product for executing such a method.

Claims

1. A method for controlling a robot device having a movable manipulator or effector comprising: providing a database storing medical injury parameters, the medical injury parameters being derived to represent a functional relationship between dynamics of the robot device, geometric primitives of the robot device and biomechanical injury data acquired from results of collision tests; monitoring a speed or direction of movement of the manipulator or effector to determine dynamics of the robot device and comparing the dynamics and geometric primitives of the robot device with the stored medical injury parameters, and adapting the speed or direction as function of the comparison of the dynamics and geometric primitives of the robot device with the stored medical injury parameters to prevent the speed or direction of movement of the manipulator or effector from approaching a value of the medical injury parameters associated with causing injuries to a human above a certain predefined degree.

2. The method as recited in claim 1 wherein the monitoring is a function of includes comparing an impact mass, an impact velocity or an impact-contact geometry of the manipulator or effector with the stored medical injury parameters.

3. The method as recited in claim 1 wherein the monitoring includes comparing an anticipated impact mass, impact velocity or impact-contact geometry of at least one prescribed relevant point of the manipulator or effector with the stored medical injury parameters.

4. The method as recited in claim 1 wherein the stored medical injury parameters include characteristic values, the characteristic values reflecting a relationship between an impact mass, an impact velocity or an impact-contact geometry of the manipulator or effector, and the medical injury parameters.

5. The method as recited in claim 4 wherein the characteristic values are depicted in mass-velocity diagrams for different contact geometries and different types of injury.

6. The method as recited in claim 1 wherein the medical injury parameters are obtained from a memory unit.

7. The method as recited in claim 1 wherein the monitoring includes using at least one adjustable threshold value.

8. The method as recited in claim 1 wherein the method is carried out in real time.

9. The method as recited in claim 1 wherein the method is carried out on a level of commands or on a level of measured values.

10. A non-transitory computer readable medium including a computer program product downloadable into a computing unit and comprising software code segments, the method as recited in claim 1 being carried out with the software code segments when the product is running on the computing unit, the computing unit controlling the robot device in accordance with the software code segments.

11. The method as recited in claim 1 wherein the relationship is provided as a characteristic curve.

12. The method as recited in claim 1 wherein the relationship is given as:
v.sub.max(m)=reg.lim[c.sub.1(i,a.sub.i)*m+c.sub.2(i,a.sub.i)v.sub.1,v.sub.2], where: c.sub.1(i, a.sub.i) is a coefficient of the safety characteristic curves, c.sub.2(i, a.sub.i) is a coefficient for the primitive I, v.sub.1 denotes a minimally permissible velocity, v.sub.2 denotes a maximally permissible velocity.

13. The method as recited in claim 1 wherein each of the relevant points POI is associated with the following information: a relative pose .sup.ObjT.sub.POI in relation to the primitive object reference coordinate system, a set of geometric parameters PARAMS that represent a surface primitive SURFACE, an identifier SC-TYPE for the type of the provided relationship between a scalar mass m, a maximum velocity v.sub.max(m) depending on the mass m and a medical injury for each of the primitive objects and for a certain part of a human body, and a set of coefficients COEFF that describe the corresponding relationship.

14. The method as recited in claim 13 wherein the information associated with the relevant points POI is provided in a database.

15. A robot device comprising: a movable manipulator or effector; and a control unit having a computing unit and a memory unit, the memory unit including a database storing medical injury parameters, the medical injury parameters being derived to represent a functional relationship between dynamics of the robot device, geometric primitives of the robot device and biomechanical injury data acquired from results of collision tests, the control unit configured for monitoring a speed or direction of movement of the manipulator or effector to determine dynamics of the robot device and comparing the dynamics and geometric primitives of the robot device with the stored medical injury parameters, and adapting the speed or direction as function of the comparison of the dynamics and geometric primitives of the robot device with the stored medical injury parameters to prevent the speed or direction of movement of the manipulator or effector from approaching a value of the medical injury parameters associated with causing injuries to a human above a certain predefined degree.

16. A method for controlling a robot device having a movable manipulator or effector comprising monitoring a speed or direction of the movement of the manipulator or effector depending on medical injury parameters and on dynamics of the robot, and adapting the speed or direction as a function of the monitoring, wherein the manipulator or effector is characterized by a group of geometrically coupled primitive objects and wherein a set of relevant points POI is defined for each of the primitive objects, comprising the steps of: providing a relationship between a scalar mass m, a maximum velocity v.sub.max(m) depending on the mass m and a medical injury for each of the primitive objects and for a certain part of a human body; providing a target movement for the manipulator or effector; depending on the target movement of the manipulator or effector, determining, for each of the relevant points POI of the primitive objects, a reflected mass m.sub.u of the manipulator or effector; depending on the target movement of the manipulator or effector and on the determined reflected masses m.sub.u and based on the relationship provided, determining a maximum velocity v.sub.max(m.sub.u) for each of the relevant points POI; and comparing, for each of the relevant points POI, a POI target velocity, which result from the target movement of the manipulator or effector, and the determined maximum velocity v.sub.max(m.sub.u) dedicated to the respective POI, and in case a POI target velocity exceeds the dedicated maximum velocity v.sub.max(m.sub.u), preventing the respective POI target velocity from exceeding the relevant maximum velocity v.sub.max(m.sub.u).

17. A robot device comprising: a movable manipulator or effector; and a control unit having a computing unit and a memory unit, the computing unit monitoring a speed or direction of the movement of the manipulator or effector depending on medical injury parameters and on dynamics of the robot, and adapting the speed or direction as a function of the monitoring, wherein the manipulator or effector is characterized by a group of geometrically coupled primitive objects and wherein a set of relevant points POI is defined for each of the primitive objects, the computer unit being programmed to execute the following steps: providing a relationship between a scalar mass m, a maximum velocity v.sub.max(m) depending on the mass m and a medical injury for each of the primitive objects and for a certain part of a human body, providing a target movement for the manipulator or effector, depending on the target movement of the manipulator or effector, determining, for each of the relevant points POI of the primitive objects, a reflected mass m.sub.u of the manipulator or effector, depending on the target movement of the manipulator or effector and on the determined reflected masses m.sub.u, and based on the relationship provided, determining a maximum velocity v.sub.max(m.sub.u) for each of the relevant points POI, and comparing, for each of the relevant points POI, a POI target velocity, which result from the target movement of the manipulator or effector, and the determined maximum velocity v.sub.max(m.sub.u) dedicated to the respective POI, and in case a POI target velocity exceeds the dedicated maximum velocity v.sub.max(m.sub.u), preventing the respective POI target velocity from exceeding the relevant maximum velocity v.sub.max(m.sub.u).

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Embodiments of the invention will be described in greater detail below making reference to figures. Additional features and advantages ensue from this description. Concrete features of these embodiments can constitute general features of the invention. Features of these embodiments combined with other features can also constitute individual features of the invention.

(2) The following is shown schematically by way of an example:

(3) FIG. 1: a selection of contact primitives for soft tissue experiments;

(4) FIG. 2: the dependence of an AO classification on the impact mass and impact velocity for different contact primitives;

(5) FIG. 3: conservative safety characteristic curves for different contact primitives;

(6) FIG. 4: an end effector consisting of contact primitives and associated POIs (points of interest);

(7) FIG. 5: a robot with the end effector for carrying out experiments;

(8) FIG. 6: trajectories of a line test and of a ribbon test; and

(9) FIG. 7: a “line test” and “ribbon test” SVC (safe velocity controller) experiment.

DETAILED DESCRIPTION

(10) Introduction, approach:

(11) Within the scope of the present invention, robot systems for human-robot interaction (HRI) are to be designed safely without simply introducing generalized limits as was the case, for instance, in ISO 10218-2006. All of the results and insights presented there are of a very general nature and are not tailor-made for a specific robot design. In principle, the present results, analyses and control algorithms can be useful for any robot. To begin with, it is necessary to understand the influence of a generic collision between a robot and a human. To be more precise, the question arises as to what the relationship is between the impact mass, the impact velocity and the contact geometry as well as the injuries that occur. This information can then be used to develop a control algorithm that utilizes this knowledge in such a way that the concrete significance of what constitutes a safe interaction is transmitted on a very low algorithmic level to the robot. This approach is especially useful so that a robot can move as quickly as possible while ensuring human safety. Fundamentally speaking, this provides an answer to the question, “How fast can I move without hurting someone?”. Before the aspect of soft-tissue injuries in robotics can be understood, it is first necessary to generate sufficient biomechanical injury data so that sufficiently meaningful safety limits can be formulated. Since, until now, neither biomechanics nor forensic medicine has actually focused on minor injuries, there is a need for adequate collision tests to be conducted in order to ascertain the relationship between “input” robot parameters, reflected inertia, velocity and impact geometry, and to acquire information about the resulting injury. Since it is not possible to test every conceivable contact geometry, sufficiently informative relevant primitives are determined in a first step. Subsequently, numerous drop tests are carried out with fresh pig-abdominal wall specimens employing different masses and velocities for specific primitives. The analyses can be expanded to other parts of the body. However, the experiments so far have already generated an enormous amount of data so that yet more experiments might actually obscure the essence of the work. For this reason, one single body region will be examined here. The injury that might have been caused is then medically assessed according to the following scheme:

(12) 1) direct medical examination of the impact area

(13) 2) preparation and injury analysis

(14) 3) histopathological examination

(15) Using the above-mentioned AO classification, the observed injury is then classified into standardized injury classes. Here, three selected primitives were evaluated on the basis of hundreds of drop tests. Here, the focus will instead be concentrated on developing an approach rather than on merely generating an extremely large volume of raw data. Data for more primitives and parts of the body can still be accumulated. In order for the functional relationship—robot parameter.fwdarw.injury—to be rendered useable for robot regulation and control, risk curves are derived for the given primitives. This makes it possible to derive a simple and intuitive representation of the relationship—robot parameter.fwdarw.injury—which can then be stored in a real-time database. The acquired injury knowledge is now available in a real time-capable structure and it can be integrated into a velocity controller. The proposed controller takes into account the reflected dynamics of the robot at relevant structure points as well as its velocity and surface characteristics. The framework elicits a safe collision behavior for a robot in case of an unanticipated collision with a human. It should also be mentioned that previous work on blunt impact analyses can be seamlessly integrated into this algorithm.

(16) Medical Evaluation:

(17) The relevant medical conventions as well as the drop-test experiments carried out will be described below. Subsequently, injuries that were observed during the experiments are evaluated medically. The medical analysis is carried out in three phases:

(18) 1) direct observation after the drop test

(19) 2) macroscopic patho-anatomical analysis

(20) 3) microscopic patho-anatomical analysis

(21) All of the phases are explained in detail. However, in order for acquiring a better understanding of the results of the observation, the drop-test experiments will be presented first. A detailed description of the protocol is presented in the publication by S. Haddadin, S. Haddadin, A. Khoury, T. Rokahr, S. Parusel, R. Burgkart, A. Bicchi, and A. Albu-Schïffer, “On making robots understand safety: Embedding injury knowledge into control”, Int. J. of Robotics Research, 2012.

(22) Drop-Test Experiments

(23) For the experimental injury analysis, a protocol was chosen that is based on the principle of a free fall. The set-up comprises a force sensor as well as two acceleration sensors that measure the acceleration of the carriage and of the impactor and that are employed to determine the impact velocity. The maximum pressure that occurs is measured using a pressure-indicating film.

(24) Since there is a large selection of possible contact geometries for such tests, a representative selection of tools and classes was drawn up. FIG. 1 shows a selection of contact primitives for soft tissue experiments. The circled primitives are used in the test series and are designated here as a small sphere 100, a large sphere 102 and a wedge 104. The specific design is based on typical geometric primitives from industrial processes.

(25) The small sphere 100 has a radius R=5 mm and a mass of 2.1 kg. The large sphere 102 has a radius R=12.5 mm and a mass of 2.2 kg. The wedge 104 has a wedge angle of 45° with a fillet radius r=0.2 mm, a width L=200 mm and a mass of 2.7 kg.

(26) The selected contact geometries already cover numerous industrial grippers or objects that are to be gripped. The analysis methodology employed for the medical evaluation of the observed injuries will be explained below.

(27) Analysis Methodology:

(28) 1) AO classification and macroscopic analysis: in medicine, minor injuries are generally treated as secondary injuries that accompany fractures. In this context, the AO classification of the “Arbeitsgemeinschaft für Osteosynthesefragen” [Association for the Study of Internal Fixation] is one of the most important internationally. In this context, reference is made to the publication by S. Haddadin, S. Haddadin, A. Khoury, T. Rokahr, S. Parusel, R. Burgkart, A. Bicchi, and A. Albu-Schïffer, “On making robots understand safety: Embedding injury knowledge into control”, Int. J. of Robotics Research, 2012. This classification is at times also known as the Müller classification after the Swiss surgeon and pioneer of orthopedic surgery, Maurice Edmond Müller. In the English-speaking world, it is often referred to as ASIF (Association for the Study of Internal Fixation). This classification aims at establishing a description of fractures of the human skeleton that is uniform and unambiguous worldwide. Moreover, there is a subgroup in the AO classification that deals with concomitant injuries such as, for instance, skin and soft-tissue injuries (muscles, ligaments, tendons, nerves and vessels). This subgroup was chosen for purposes of allowing a precise and objective description of the experimentally induced injuries. The goal of the AO classification is to improve communication among physicians and to improve systematic documentation and research. Consequently, sensible therapeutic approaches are to be selected that are based on appropriate, clearly described and easily accessible data, thus forming the foundation for today's evidence-based medicine. The groupings of the AO classification used here are:

(29) 1) skin injuries (I)

(30) 2) muscle and tendon injuries (MT) and

(31) 3) vascular and nerve injuries (NV)

(32) Moreover, these classes are divided as follows:

(33) Skin Injury in Case of a Closed Fracture:

(34) IC1: no skin injury

(35) IC2: contusion without breaking of the skin

(36) IC3: circumscribed décollement

(37) IC4: extensive, closed décollement

(38) (IC5: necrosis due to deep contusion)

(39) Open Skin Injury:

(40) (IO1: skin puncture from the inside to the outside)

(41) IO2: skin puncture from the outside <5 cm with contused edges

(42) IO3: skin lesion >5 cm, circumscribed décollement with edge contusion

(43) IO4: skin loss, deep contusion, abrasion

(44) IO5: (extensive décollement)

(45) Muscle and Tendon Injuries:

(46) MT1: no injury

(47) MT2: circumscribed muscle injury (limited to one muscular compartment)

(48) MT3: extensive muscle injury (in two or more muscular compartments)

(49) (MT4: avulsion or loss of entire muscular compartments, severed tendons)

(50) (MT5: compartment or crush syndrome)

(51) Neurovascular Injuries:

(52) NV1: no injury

(53) NV2: isolated nerve lesion

(54) NV3: circumscribed vascular injury

(55) NV4: combined neurovascular injury

(56) (NV5: subtotal or total amputation)

(57) Here, IC2 is considered to be the appropriate limit value and it is designated as an indicator of the “key impact”. In addition to the evaluation on the basis of the AO classification, the widths, lengths and depths of the lesions that occur are manually measured using calipers. For documentation purposes, photos are made of each specimen before and after each test series. A preliminary observation and evaluation were made after each impact. Upon conclusion of a test series, the specimen is removed from the test set-up in order to undergo a thorough examination. First, the skin surface is examined and classified according to IC1-5 or IO2-5. If no obvious breaking of the skin can be ascertained, specimens of 1 cm.sup.3 are removed and fixed in formalin for purposes of later microscopic examinations. The objective of the microscopic analysis is to make a detailed distinction, which in certain cases would not be possible only with a macroscopic analysis. Abdominal wall tissue was selected as the first test tissue because, with it, it is relatively easy to carry out drop tests under equivalent impact conditions.

(58) Results:

(59) It should be pointed out that the evaluation presented here is a summary of the observations which are especially intended to serve for the design of the controller described below rather than to serve for the interpretation of the results. FIG. 2 summarizes the risk characteristic curves obtained from the tests and shows the dependence of the AO classification on the impact mass and impact velocity for various contact primitives. The first column with diagrams 200, 202, 204 shows the dependence of the AO classification on the impact mass and impact velocity for a wedge (see FIG. 1 and the appertaining description), the center column with diagrams 206, 208, 210 shows the dependence of the AO classification on the impact mass and impact velocity for a large sphere (see FIG. 1 and the appertaining description), while the right-hand column with diagrams 212, 214, 216 shows the dependence of the AO classification on the impact mass and impact velocity for a small sphere (see FIG. 1 and the appertaining description). The top row with diagrams 200, 206, 212 shows the results for closed skin injuries, the center row with diagrams 202, 208, 214 depicts the results for muscle and tendon injuries while the bottom row with diagrams 204, 210, 216 shows the results for neurovascular injuries.

(60) As already mentioned, a key impact is ascribed to each impactor, each velocity as well as each mass. The key impact is the maximally permissible injury that is allowed to occur. It is defined here as contusion. Of course, this definition is not sufficient if the skin remains completely intact but the tissue underneath it is injured. This is particularly the case when nerves and arteries are involved. The third class of soft-tissue injuries, neurovascular damage, is already possible in cases of penetrating muscle injuries since major neurovascular structures are located underneath the muscles. Consequently, the key impact is selected as a function of the estimated injury to the human being, namely, it has to be totally reversible (restitutio ad integrum), that is to say, it must not leave any permanent damage. If the results are not unambiguous, the key impact is determined using the most conservative interpretation. The results of the drop test are then integrated into the real-time robot control procedure as will be described below. Each medical result is integrated into the injury database, which will be elaborated upon below.

(61) FIG. 2 shows the relationships between mass, velocity and injury for 276 drop tests. Each impactor, wedge, large sphere, small sphere is associated with a column depicting the skin injuries, muscle and tendon injuries as well as neurovascular injuries. The severity of the tissue damage (ranging from one to five as set forth in the AO classification) is represented by rectangular grayscale fields. White areas stand for impacts that do not cause injury (IC1, MT1 or NV1). Black areas represent the most severe injury possible. Here, it should be noted that in diagram 212, closed as well as open skin injuries are shown together in one graph. Here, black areas designate open skin injuries and non-closed skin injuries with by IC5 (necrosis due to deep contusion).

(62) Skin injuries caused by the wedge-shaped impactor are limited to contusions and closed décollements. Only masses >8 kg and impact velocities of at least 3.0 m/s lead to more severe contusions and décollements. The most severe injuries for the large sphere are small avulsions≈10 mm.sup.2. At velocities below 2 m/s, the large sphere only causes minor skin damage. Neither the large sphere nor the wedge were able to completely pierce the skin, which was verified macroscopically as well as microscopically. In contrast, the small sphere pierces the skin at relatively low velocities and masses. Safety in case of abdominal impacts can only be guaranteed for velocities <2 m/s and masses <6 kg. The wedge impacts caused only negligible muscle injuries in all of the tests up to 1.5 m/s. Even above this velocity, most of the injuries could be considered as harmless.

(63) Consequently, the wedge entailed a relatively wide safety margin for muscle tissue. Muscle injuries caused by the larger sphere at >8 kg can be seen as tolerable up to 1.5 m/s. The results for the small sphere show similar degrees of severity of damage to the muscle tissue. The maximum velocity should not exceed 1.5 m/s. In the case of greater masses, the velocity should be limited to 1.0 m/s. As already mentioned above, neurovascular injuries are selected as a function of the muscle penetration. However, no examination of this type of injury was carried out here since it is not possible to conduct an adequate evaluation using non-living tissue.

(64) Mention should be made of the fact that all of the experiments conducted entail certain conditions that deviate from real human-robot collisions. These are primarily due to the use of non-living tissue. In comparison to living tissue, non-living tissue lacks several properties such as muscle tone, pre-stretching of the skin and, of course, the possibility of evading the impact. Moreover, it is not possible to examine functional damage such as, for example, arterial/venous hemorrhaging, pain or neurological failure. Nevertheless, the experiments carried out and their results tend more to reflect a worst-case scenario, rather than focusing on ameliorating possible outcomes. Safety curves will be derived below which are suitable for a real-time evaluation so that, on the basis of its current state (reflected inertia, instantaneous speed and surface), a robot is capable of drawing conclusions about its potential for causing injury in the case of unanticipated collisions. It is shown how this representation can be used in a closed control loop so that the robot does not exceed the applicable medical limit values.

(65) Knowledge-Based Real-Time Control:

(66) Safety Characteristic Curves for Robot Control:

(67) The objective of the present invention is to understand how soft tissue responds under different impact test conditions, in order to find suitable model parameters that make it possible to predict the occurrence of a particular injury and then to integrate this knowledge into a control unit for safe robot speeds. Even though additional experimental data is helpful for a complete understanding of the mapping—(mass, velocity, geometry).fwdarw.injury—such a complete characterization of soft tissue is certainly not necessary for robot applications: the prediction of velocity limits for very small masses (<1 kg) is not important for two reasons:

(68) Lower limit: the reflected inertia for robots that are typically employed in interactive applications is considerably greater, especially if the robot is equipped with a gripper/hand and/or tools.

(69) Upper limit: a robot that is present in the immediate vicinity of a human being and/or that is cooperating with said human during work should certainly not exceed 4 to 5 m/s. It has already been demonstrated that ≈2 m/s is a reasonable maximum limit speed. This limit was derived from experiments on blunt impact to the head.

(70) Moreover, an evaluation of speed limits for large masses (>20 kg) is not of crucial interest in service robotics since robots that are to interact safely with humans typically have a lightweight construction and therefore have a reflected inertia within the range from 1 kg to 15 kg and, in the case of very large reflected inertias (for example, in the vicinity of singularities), it does not make sense to reduce the speed below a certain value or to even stop the movement. Singularities are not taken into consideration at this juncture. They have to be analyzed in a different manner. Whereas the reflected mass approaches infinity, the velocity moves towards zero, that is to say, the kinetic energy is limited.

(71) Finally, it seems appropriate to establish a maximum permissible velocity for small masses (e.g. 4.5 m/s) and to define a minimum velocity limit (e.g. 0.1 m/s) in order to prevent the robot from stopping in the vicinity of singularities. This is why all requisite information is acquired from the test results.

(72) The last unresolved question is which representation reflects the limits in the mass-velocity characteristic curves. Initially, one would select a description in terms of physical quantities such as kinetic energy, force of contact or momentum. However, since a medical evaluation on the basis of the AO classification is available, the prediction of injuries does not require a physical model, but rather, it is exclusively data-driven. Therefore, in view of the complexity of human injury mechanisms, more consistent results can be obtained than with a model-based approach, which requires validation and potentially entails greater imprecisions. Consequently, all of the measurements of physical quantities during an experiment can be seen as supplementary information. However, they are not required for the mapping—(mass, velocity, geometry)—injury—(of a given part of the body). Since the mass and the velocity were associated with the “key impact” for experiments on the abdomen, the resulting safety curves for the abdomen in the experiments are simply three regression curves in the (mass/velocity) plane for a given impact primitive. The maximally permissible velocity can be expressed as follows:
v.sub.max(m)=reg.lim[c.sub.1(i,a.sub.i)m+c.sub.2(i,a.sub.i)v.sub.1,v.sub.2],  (1)
with the coefficient of the safety characteristic curves c.sub.1(i, a.sub.i)<0 and c.sub.2(i, a.sub.i) for the primitive i. The parameters v.sub.1, v.sub.2 denote the minimally and maximally permissible velocities. It should be emphasized once again that no force sensor is needed in order to delimit the safety characteristic curves for the robot control. All that is needed is knowledge about the velocity of a varying mass at which a medically detectable injury occurs (that is to say, the mapping of the mass, velocity and geometry onto the medically observable injury). The sensor data, in contrast, can be employed for applications that explicitly require this such as, for instance, force-controlled tasks for predicting contact forces.

(73) FIG. 3 shows conservative safety characteristic curves for the contact primitives in the form of a small sphere in diagram 300, a large sphere in diagram 302 and a 45°-wedge in diagram 304, which are used in the real-time control procedure for the abdominal area.

(74) The maximum speeds are evaluated within the range from 0.1 m/s to 4.5 m/s. The resulting characteristic curves are subsequently shifted conservatively so that all of the data points lie above the applicable limit. These characteristic curves form the basis for integrating interpretable knowledge about injury mechanisms into real-time robot control procedures, which will be described below.

(75) Injury Database:

(76) The described results of the impact tests yield safety curves that relate the maximum velocity and mass to the injury for a certain primitive and for a certain part of the body; see FIG. 3. Here, the objective is to make it possible for a robot to utilize this knowledge for purposes of limiting the velocity of moving parts in such a way that accidental collisions do not exceed a given limit value as set forth in the AO classification, here IC2. In order to attain the highest possible velocities while complying with a specific safety limitation, the mapping (mass, velocity) vis-à-vis observed injury has to be made available online. Such an injury database stores the coefficients of the safety characteristic curves (1) for each known contact primitive so that these can be utilized in real time to adapt the speed. In actual practice, end effectors or relevant robot structures cannot be represented by only one single contact primitive. These are rather complex geometric objects with varying characteristics. Owing to this large number of characteristics, it is naturally not practical to treat every single end effector/robot separately. For purposes of developing a methodical approach, end effectors are formally broken down into groups of geometrically coupled primitives.

(77) The entire geometric structure of the broken-down end effectors can be expressed in terms of relative transformation matrices .sup.EET.sub.Obj between the end effector and the primitive-object reference coordinate system. The individual geometric, dynamic (obtained from a CAD model, dynamic identification, learning, etc.) and safety properties adequately describe the robot shell. It should be noted that no algorithm is given here for the automatic establishment of this relationship. Each relevant point that is to be monitored is defined as a point of interest (POI). Each primitive object consists of a set of POIs, the position of the center of gravity .sup.Objx.sub.COG, the mass m, the inertia sensor I and the relative position to the end effector .sup.EET.sub.Obj.

(78) The following properties are associated with each POI. First of all, its relative pose in relation to the primitive-object reference coordinate system .sup.ObjT.sub.POI. Secondly, a set of geometric parameters PARAMS that represent the surface primitive SURFACE. Thirdly, an identifier SC-TYPE for the type of safety curve, here a limited regression. Fourthly, a set of coefficients COEFF that describe the corresponding safety curve. Together, COEFF, PARAMS and SC-TYPE form the PRIMITIVE structure for each POI.

(79) The resulting database is formally represented as follows:
SoEEs={SoObjects.sup.k×{.sup.EET.sub.Obj}.sup.k}
.sup.EET.sub.ObjεSE(3)
SoObjects={POI.sup.m×.sup.3×.sup.+×.sup.3×3}
POI={.sup.ObjT.sub.POI×PRIMITIVE}
.sup.ObjT.sub.POIεSE(3)
PRIMITIVE={COEFF×PARAMS×SC−TYPE}
COEFFεSoC(SC−TYPE)
PARAMSεSoP(SURFACE)   (2)

(80) SoEES is the set of end effectors, while SoObjects is the set of primitive objects.

(81) Since the injury database contains only relationships between the scalar mass, the velocity and the injury, the instantaneous reflected mass of a given POI and its target velocity have to be calculated for a given movement. Thus, the stored information can be employed in such a way that the target speed is scaled so as to be reduced under (1) in terms of the potential risk of injury.

(82) Dynamics-Based Speed Scaling in Real Time:

(83) 1) Reflected mass at the POI: the dynamics of a rigid robot in the articulation space are described by
M(q){umlaut over (q)}+C(q,{dot over (q)}){dot over (q)}±g(q)=τ,  (3)
wherein qεR.sup.n stands for the vector of the articulation angle, M(q)εR.sup.n×n stands for the inertia matrix, C(q,{dot over (q)}) stands for the centrifugal and Coriolis matrix, g(q) stands for the gravity vector and τ stands for the articulation torque.

(84) The relationship between the articulation velocities and the Cartesian velocities is expressed by {dot over (x)}=J(q){dot over (q)}, wherein J(q)εR.sup.6×n is the corresponding Jacobian matrix. M(q) and the Cartesian kinetic energy matrix Λ(x) are interrelated as follows:
Λ(x)=(J(q)M(q).sup.−1J.sup.T(q).sup.−1.  (4)

(85) In this context, reference is made to the publication by O. Khatib, “Inertial properties in robotic manipulation: an object-level framework” Int. J. Robotics Research, vol. 14, no. 1, pp. 19-36, 1995.

(86) On the basis of a breakdown of the kinetic energy matrix, the following inverse is obtained

(87) $\begin{array}{cc}{\Lambda }^{-1}\left(q\right)=\left[\begin{array}{cc}{\Lambda }_v^{-1}\left(q\right)& {\stackrel{\_}{\Lambda }}_{vw}\left(q\right)\\ {\stackrel{\_}{\Lambda }}_{vw}^T\left(q\right)& {\Lambda }_w^{-1}\left(q\right)\end{array}\right].& \left(5\right)\end{array}$

(88) A scalar quantity is obtained which, in view of a force in the u-direction, represents the observable mass on the end effector, whereby u is a unit vector. This quantity is called the reflected robot inertia in the u-direction. It should be noted that the Jacobian matrix has to be the corresponding center-of-gravity Jacobi or, otherwise, the complete inverse from Equation (5) has to be utilized instead of only the translatory portion.
m.sub.u=[u.sup.TΛ.sub.v.sup.−1(q)u].sup.−1  (6)

(89) In accordance with the results of the drop test, m.sub.u is needed in order determine the maximally permissible Cartesian velocity in the u-direction that satisfies the safety characteristic curve; see FIG. 3.

(90) Subsequently, the maximally permissible velocity of a Cartesian point is derived which constitutes the foundation for the calculation of the safe robot speed.

(91) 2) Injury-based speed scaling: the following scheme yields a safe speed for any desired POI. It should be noted that all of the indices were left out for the sake of clarity. Moreover, the dependence of q has been left out wherever it is obvious. The base coordinate system of the robot is denoted as {0}, while that of the end effector (operational frame) is denoted as {EE}.
1) Evaluate the unit vector u that is perpendicular to the object surface of the POI (direction of the z-axis in POI coordinates),
u=.sup.0R.sub.POIz.sub.POI  (7)
wherein .sup.0R.sub.POI stands for the rotation matrix from POI coordinates in {0} coordinates.
2) Calculate .sup.0v.sub.POI on the basis of the end effector target velocity .sup.0v.sub.EE.sub.d=[.sup.0{dot over (x)}.sub.EE.sub.d.sup.0 ω.sub.EE.sub.d].sup.T.

(92) $\begin{array}{cc}{}^0{v}_{\mathrm{POI}}=\left[\begin{array}{c}{}^0{\stackrel{.}{x}}_{\mathrm{POI}}\\ {}^0{w}_{\mathrm{POI}}\end{array}\right]=\left[\begin{array}{cc}{I}_3& -{\hat{p}}_{\mathrm{POI}}\\ 0_3& I_3\end{array}\right]\left[\begin{array}{c}{}^0{\stackrel{.}{x}}_{{\mathrm{EE}}_d}\\ {}^0{w}_{{\mathrm{EE}}_d}\end{array}\right],& \left(8\right)\end{array}$
wherein P.sub.POI=[p.sub.x,POI p.sub.y,POI p.sub.z,POI].sup.T stands for the position vector of the POI in relation to {EE}.
The matrix {circumflex over (p)}.sub.POI is

(93) $\begin{array}{cc}{\hat{p}}_{\mathrm{POI}}=\left[\begin{array}{ccc}0& -p_{z,\mathrm{POI}}& p_{y,\mathrm{POI}}\\ p_{z,\mathrm{POI}}& 0& -p_{x,\mathrm{POI}}\\ -p_{y,\mathrm{POI}}& p_{x,\mathrm{POI}}& 0\end{array}\right].& \left(9\right)\end{array}$
3) Evaluation of the inverses of the Cartesian kinetic energy matrix at the POI:
Λ.sub.v,POI.sup.−1=J.sub.v,POIM.sup.−1J.sub.v,POI.sup.T  (10)
4) Reflected inertia in the u-direction through (6):
m.sub.POI=1/(u.sup.TΛ.sub.v,POI.sup.−1u)  (11)
5) Derivation of the maximum velocity v.sub.max for m.sub.POI using the safety functions.
v.sub.max(m)=reg.lim[c.sub.1(i,a.sub.i)m+c.sub.2(i,a.sub.i),v.sub.1,v.sub.2],  (12)
6) Comparison of v.sub.max to the projection of .sup.0v.sub.POI in the u-direction v.sub.loc: if ∥v.sub.loc∥≦∥v.sub.max∥, the target velocity is retained. If ∥v.sub.loc∥>∥v.sub.max∥, the new velocity is .sup.0v′.sub.POI

(94) $0_{v_{\mathrm{POI}}^{\prime }}=0_{v_{\mathrm{POI}}}\frac{.Math.v_{\max }.Math.}{.Math.v_{\mathrm{ort}}.Math.}.$
7) Finally, the new velocity .sup.0v.sub.EE of the end effector is

(95) $\begin{array}{cc}{0}_{v_{\mathrm{EE}}}=\left[\begin{array}{c}{}^0{\stackrel{.}{x}}_{\mathrm{EE}}\\ {}^0{w}_{\mathrm{EE}}\end{array}\right]=\left[\begin{array}{cc}{I}_3& {\hat{p}}_{\mathrm{POI}}\\ 0_3& I_3\end{array}\right]\left[\begin{array}{c}{}^0{\stackrel{.}{x}}_{\mathrm{POI}}^{\prime }\\ {}^0{w}_{\mathrm{POI}}^{\prime }\end{array}\right].& \left(13\right)\end{array}$

(96) This procedure is repeated for each POI. The most conservative velocity .sup.0v.sub.EE is selected as .sup.0v′.sub.EE.

(97) A number of experiments will be discussed below to explain the system behavior in simple movements going from position A to position B. For this purpose, an articulated manipulator is fitted with an end effector made up of the primitives that had been previously used in the drop experiments.

(98) Experiment:

(99) FIG. 4 shows an end effector 400 consisting of contact primitives and the associated POIs 402, 404, 406, 408. FIG. 5 shows a robot 500 with the end effector 400 for carrying out experiments. In order to show the performance capability of the developed injury-based algorithm, the lightweight robot 500 is equipped with the end effector 400 consisting of the geometric primitives that were employed for the drop tests; see FIG. 4 and FIG. 5. For this end effector 400, four POIs 402, 404, 406, 408 were selected, namely, two (402, 408) at the tips of the spheres 410, 412 and two (404, 406) on the wedge 414. Two POIs 404, 406 have to be used for the wedge 414 since the width of the wedge 414 has a significant influence during a rotational movement. If, for example, two POIs 404, 406 are selected on the edge of wedge, then one of them is, without a doubt, the fastest point of the primitive.

(100) The experiments conducted, however, cannot be utilized as a source of injury knowledge for the corners of the wedge 414 since the applicable analysis still has to be carried out. As a compromise, the two POIs 404, 406 are assigned to the distal ends of the primitive, whereby the velocity difference compared to that of the corners is negligible. Since the robot 500 was developed explicitly for a very delicate and safe interaction, its lightweight design is one of its most important properties.

(101) Since the robot 500 with the appertaining tools is not capable of generating potentially “unsafe” movements in view of its maximum speed and inertia properties, the safety characteristic curves are shifted to such an extent that the effect of the safe velocity controller (SVC) is also evident for this manipulator; scaling factor of 0.2.

(102) Experiments for two different types of movement were conducted. FIG. 6 shows the trajectories of a line test 600 and of a ribbon test 602. In the line test 600, the end effector 400 moves laterally (.sup.0y-direction) between two positions. In the ribbon test 602, the robot 400 moves crosswise to four different positions, combining vertical and horizontal movements in this process.

(103) FIG. 7 shows an SVC experiment comprising a “line test” and a “ribbon test”. The diagram 700 depicts the results of the line test 600. The end effector 400 is supposed to move at a velocity of 1.5 m/s between two target configurations. The SVC, however, limits the velocity as a function of the reflected inertia and the direction of movement of the robot 400. During the movement in the positive .sup.0y-direction, the POI 402 is the critical POI, while the POI 408 is the critical POI in the negative direction.

(104) Since the safety characteristic curve associated with POI 402 is more restrictive than that of POI 408 (the small sphere is more dangerous than the large sphere), the SVC reduces the maximum velocity more markedly in the positive .sup.0y-direction. The results of the ribbon test 602 are depicted in the diagram 702.

(105) Since the coordinate system of the end effector ({0} coordinate) is rotated with respect to the Cartesian world coordinate system ({W} coordinate system), the velocity for all three dimensions is presented. Lateral movements are represented as the .sup.0y-direction. Vertical movements have entries in the .sup.0x-direction as well as in the .sup.0z-direction.

(106) Accordingly, the segments 1-2 and 3-4 in FIG. 7 (upwards movement of the end effector 400) correspond to the positive .sup.0x-direction and to the negative .sup.0z-direction in FIG. 7. In contrast to this, the segments 2-3 and 4-1 correspond to the negative .sup.0x-movement and to the positive .sup.0z-movements. For lateral movements, the segments 2-3 and 4-1 denote negative and positive .sup.0y-movements. In FIG. 7, it can be seen that the SVC restricts the 4-1 movement more markedly than the 2-3 movement. This can be explained by the fact that, in this direction, POI 402, which is associated with the small sphere, is relevant. In spite of this fact, POI 404 and POI 406 contribute to limiting the velocity. Due to the symmetry of the end effector 400, they equally influence the velocity limits along 4-1 and 2-3 to equal extents. Finally, it can be seen that the velocity in upwards movements (1-2 and 3-4) has not yet been limited by the controller. This can be explained by the fact that the POIs 402, 404, 406, 408 were assigned exclusively frontally and laterally, and none at the rear part of the end effector 400.

SUMMARY

(107) Drop test experiments are presented here involving varying masses, velocities and geometries on pig-abdomen wall specimens. The purpose of the study is to generate fundamental injury data that does not yet exist, either in biomechanics or in forensics. Moreover, an approach is being put forward to utilize a medical evaluation and classification in order to generate an adequate representation of injury knowledge so as to allow further algorithmic processing. The basis of the evaluation is the so-called AO classification. Generally speaking, this allows an objective description of soft-tissue damage on the basis of medical observations. IC2 (contusion without opening of the skin) is selected as the maximally permissible injury limit. Thirdly, risk graphs are drawn up that represent a “safe” speed, at a given instantaneous configuration, mass, surface geometry and part of the body involved in the collision. Moreover, a real-time injury database architecture is being put forward that makes the generated results accessible in real time. Finally, a real-time controller is being designed and experimentally verified which limits the end-effector velocity on the basis of the reflected inertia and the geometric surface properties of the end effector. This device ensures that, thanks to the utilization of the knowledge from the injury database, a possible collision with a human (human abdomen) cannot cause injuries above a certain predefined degree. Finally, it should be pointed out that existing work on blunt human-robot collisions can be integrated very easily into this architecture and algorithm, which makes the present approach very generic.

REFERENCE NUMERALS

(108) 100 small sphere 102 large sphere 104 wedge 200 diagram 202 diagram 204 diagram 206 diagram 208 diagram 210 diagram 212 diagram 214 diagram 216 diagram 300 diagram 302 diagram 304 diagram 400 end effector 402 POI 404 POI 406 POI 408 POI 410 sphere 412 sphere 414 wedge 500 robot 600 line test 602 ribbon test 700 diagram 702 diagram