Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
ROBOT CONTROL METHOD
Document Type and Number:
WIPO Patent Application WO/2019/025255
Kind Code:
A1
Abstract:
A method for controlling a robot (1) comprises the steps of: a) deciding (S2) whether there is a non-permanent object (18, 19, 20) in the vicinity of the robot (1); b) if there is a non-permanent object (18, 19, 20), deciding (S5, S6, S11,S12) whether the object qualifies for extended protection or not, and c) defining a safety zone (21) around the object (18, 19, 20) which the robot (1) must not enter or in which a maximum allowed speed of the robot (1) is less than outside said zone, wherein the safety zone (21) extends to a greater distance from said object (18, 19, 20) if it qualifies for extended protection than if it does not.

Inventors:
LEHMENT, Dr. Ing. Nicolas (Brudermühlstr. 55, München, 81371, DE)
DECKER, Dr. rer.nat. Andreas (Evenaristr. 52, Darmstadt, 64293, DE)
ROBERTS, Ph. D. Richard (Frühlingstr. 5A, Gilching, 82205, DE)
MATTHIAS, Dr. Björn (Zeuterner Str. 4, Bad Schönborn, 76669, DE)
TIAN, Jihuan (6 102, Lane 3088 Jinxiu Roa, Pudong Shanghai 4, 201204, CN)
Application Number:
EP2018/070161
Publication Date:
February 07, 2019
Filing Date:
July 25, 2018
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
ABB SCHWEIZ AG (Brown Boveri Strasse 6, 5400 Baden, 5400, CH)
International Classes:
B25J9/16; F16P3/14
Domestic Patent References:
WO2010060475A12010-06-03
Foreign References:
US7778776B22010-08-17
US20150158178A12015-06-11
US20150061880A12015-03-05
US20080021597A12008-01-24
US20110025547A12011-02-03
US20140316261A12014-10-23
Attorney, Agent or Firm:
MARKS, Frank (Wallstadter Strasse 59, Ladenburg, 68526, DE)
Download PDF:
Claims:
Claims

1. A method for controlling a robot (1) compris¬ ing the steps of:

a) deciding (S2) whether there is a non- permanent object (18, 19, 20) in the vicinity of the robot ( 1 ) ;

b) if there is a non-permanent object (18, 19, 20), deciding (S5, S6, Sll, S12) whether the object qualifies for extended protection or not, and

c) defining a safety zone around (21) the object (18, 19, 20) which the robot (1) must not enter or in which a maximum allowed speed of the robot (1) is less than outside said zone, wherein the safety zone (21) extends to a greater distance from said object (18, 19, 20) if it qualifies for extended protection than if it does not.

2. The method of claim 1 in which when an object (18, 19, 20) is detected (S5) to enter the vi¬ cinity of the robot (1), it is judged (S6) to qualify for extended protection.

3. The method of claim 1 in which the vicinity comprises a workpiece entry point (17) and an object (6) which enters the vicinity by the workpiece entry point (17) is judged (S5, 7) not to qualify for extended protection.

4. The method of claim 1, 2 or 3, wherein if the distance between two objects (19, 9) becomes less than a predetermined limit, and at least one of the objects (19) has been judged to qualify, the other object (9) is judged to qualify (Sll, S12), too.

The method of claim 4, wherein if the other object (9) did not qualify prior to the dis¬ tance becoming less than the predetermined limit, and if the distance increases again without having become less than a pre-set threshold, which may be defined as zero or as close to zero, the other object (9) is again judged not to qualify.

The method of any of the preceding claims, wherein if two objects (9, 19) come close enough to each other to become undistinguisha- ble and at least one of the objects (19) has been judged to qualify, the resulting object (22) is judged to qualify, too.

The method of any of the preceding claims, wherein if a qualifying object (18) splits into two or more objects (19, 20), each result¬ ing object is judged to qualify, too.

The method of any of the preceding claims, wherein an object (9, 20) is judged not to qualify if at least one of the following cri¬ teria is met:

no motion of the object (9, 20) is de¬ tected within a predetermined time span,

at least one detectable physical property of the object (9, 20) selected among at least size, shape, weight or surface temperature is incompatible with a human being;

the object (9, 19) is equipped with a tag (24), and data read from the tag (24) fit de¬ tectable physical properties of the object (9, 19) ; the object is declared not to qualify by an operator.

A robot system comprising a robot (1), at least one sensor (10) for surveilling a vicinity of the robot and a processor unit (5) for controlling the robot (1) based on data from the sensor (10) using the method of one of the preceding claims.

A computer program product comprising program code means which enable a computer to perform the method according to any of claims 1 to 8.

Reference Numerals

1 robot

2 base

3 arm

4 end effector

5 processor unit

6 first workpiece

7 conveyor

8 second workpiece

9 crate

10 sensor

11 wall, barrier

12 passage

13 safety zone

14 rack

15 passage

16 safety zone

17 passage

18 object

19 person

20 crate

21 safety zone

22 combined object

23 user interface

24 tag

25 radio interface

Description:
Robot control method

The present invention relates to a method for con ¬ trolling a robot and to a robot system in which the method is used.

Wherever robots are used in industry, protecting staff from injury through the robots is a prime concern. Conventionally, enclosures have been built around the robots, equipped with alarms which cause an emergency stop of the robot whenever a person opens the enclosure. Such robots can only accom ¬ plish fully automatized, highly repetitive tasks that require no human intervention at all. A direct interaction between robot and human is not possible .

In order to enable robots and humans to co-operate without being separated by an enclosure, more so- phisticated security concepts are required. One such concept is known as speed and separation monitoring (SSM), described in ISO/TS 15066:2016 - Ro ¬ bots and robotic devices - Collaborative Robots, clause 5.5.4. SSM aims to maintain a certain combi- nation of distance between human and robot and ro ¬ bot motion speed, so as to ensure that the human in the robot work space cannot reach the moving robot. To achieve this, the position of the human must be determined and the robot must adjust its speed and/or trajectory accordingly, maintaining a safe distance from the human. A sensor having the spatial resolution required for detecting the position of the human, such as a video camera or a radar system, is not per se capable of distinguishing between humans and non-human, in particular inanimate, objects. Motion as such is not a criterion, since any object newly appearing within the detection range of the sensor, can only have got there by moving or by having been moved. However, if any object detected by the sensor is assigned a safe distance which would be appropriate for a human, interaction between human and robot would still be practically impossible since the ro ¬ bot would be unable to touch any object brought in- to its reach, and the robot might easily be slowed down or even prevented from carrying out a programmed task by an object it cannot pass without violating its safe distance. It is an object of the present invention to provide a method for controlling a robot which ensures the necessary safety for humans in the vicinity of the robot while avoiding unnecessary limitations in the freedom of movement of the robot.

The object is achieved by a method for controlling a robot comprising the steps of:

a) deciding whether there is a non-permanent ob ¬ ject in the vicinity of the robot;

b) if there is a non-permanent object, deciding whether the object qualifies for extended protec ¬ tion or not, and

c) defining a safety zone around the object which the robot must not enter or in which a maximum al- lowed speed of the robot is less than outside said zone, wherein the safety zone extends to a greater distance from said object if it qualifies for ex ¬ tended protection than if it does not;

d) The safety zone is updated in real time and al ¬ ways contains and goes with the object. The safety zone also contains an additional margin surrounding the object, which takes into account the sensor system delay and the maximum speed of the object.

If an object does not qualify for extended protec- tion, the distance to which the safety zone extends may be zero; i.e. the robot may be allowed to touch or even to grip and manipulate the object.

The vicinity can be regarded as equivalent to the detection range of the sensor or of a plurality of sensors on whose outputs the decision of step a) is based .

In principle, an object which qualifies for extend- ed protection should be any human, whereas an ob ¬ ject which does not should be inanimate. However, when an object has newly entered the vicinity of the robot, it may not yet be possible to tell whether it is human or not; therefore, when an ob- ject is detected to enter the vicinity of the ro ¬ bot, it should initially be judged to qualify for extended protection, notwithstanding that this judgment may be reversed later, when more infor ¬ mation on the object has been gathered.

Workpieces also have to be brought into the vicini ¬ ty in order to be worked upon by the robot. If these were initially judged to qualify for extended protection as mentioned above, the robot would be prevented from touching them and working on them at least until the information has been gathered by which the workpiece can be judged to be inanimate, whereby the work of the robot would be slowed down considerably. In order to avoid unnecessary delays in the operation of the robot, a workpiece entry point should be defined in the vicinity of the ro- bot, so that an object which enters the vicinity by the workpiece entry point can be judged not to qualify for extended protection. In that case, of course, the vicinity of the robot should be de ¬ signed so that the workpiece entry point is inac- cessible to a human from outside the vicinity.

For a first object which was found not to qualify for extended protection (also referred to subse ¬ quently as non-qualifying object) the safety dis- tance can be set very low or even zero because the first object can safely be expected not to move un ¬ predictably and possibly hit the robot. Of course such an assumption is no longer valid if the first object is seized by a human. Therefore, if the dis- tance between the first object and a second object becomes less than a predetermined limit, and at least one of the objects has been judged to qualify for extended protection, then the other object is judged to qualify, too.

Such a change in the qualification of the first ob ¬ ject causes an abrupt increase of its safety zone, by which the robot may suddenly find itself in the safety zone without having moved. In order to avoid the need for abruptly braking the robot in such a case, it is useful to requalify the first object before the human actually has been able to seize it, i.e. at a time when the distance between the second object (which might be a human) and the first object is not yet less than a pre-set thresh ¬ old, which is defined as to characterize seizure of the first object by the human, this threshold may be defined as zero or as close to zero.

If it turns out later that the second object has passed by the first object - i.e. the distance be ¬ tween the objects increases again - without the distance actually having become less than the pre ¬ set threshold, the status of the first object can be reverted to non-qualifying.

If two objects come close enough to each other to become undistinguishable, typically because a human seizes an inanimate object or the human is partly or fully shadowed by the object in the direction of sensor detection, the combined object resulting therefrom should also be judged to qualify for ex ¬ tended protection.

If such an object splits into two or more objects, e.g. because the human puts down a previously seized object, it may be impossible, at least until enough data for a reliable judgment have been gath ¬ ered, to tell whether one of the two objects is in ¬ deed inanimate. Therefore, the qualification for extended protection should be inherited by each of the objects resulting from the split.

The decision that an object is most probably inani ¬ mate and does therefore not qualify for extended protection can be based on various criteria.

One criterion is the absence of detectable motion of the object. An advantage of this criterion is that it can be examined, albeit roughly, at no ex ¬ tra cost, using the same spatially resolving sensor as for detection of the object. Reliability of the criterion is considerably improved if a highly sen ¬ sitive sensor capable of detecting micro-motion due to respiration or heartbeat is used, as described e.g. in US 2011 025547 Al or US 2014 316261 Al .

As another criterion, it can be judged whether a physical property of the object such as size, shape, weight or surface temperature is incompati ¬ ble with a human being or not. Judgment of shape and size can also be carried out based on data from the spatially resolving sensor mentioned above. Such a judgment must take account of the fact that a detected object may comprise a human and an inan ¬ imate object carried by him, so that an object can be judged not to qualify with certainty if the shape or a dimension of the object is too small for a human to fit in.

For a judgment based on weight or surface tempera ¬ ture, additional sensors become necessary, such as weight-sensitive sensor mats covering the floor of the robot's vicinity, a pyrometer for contact-free temperature measurement, or the like.

Judgment may also be facilitated by the use of spe ¬ cific tags. Objects which are intended to be used in the vicinity of the robot, e.g. workpieces to be worked on by the robot, tools etc. can be provided with tags that are adapted to be read by a sensor of the robot system and which specify the nature of the tagged object. There can be an optical tag, e.g. a QR tag, which can be read by a camera serv ¬ ing as the spatially resolving sensor; an RFID tag has the advantage that it can be read regardless of whether it faces a sensor antenna or not. The data of the tag should not be used as the sole indicator of the nature of an object; it should be checked whether the data read from the tag fit de- tectable physical properties of the detected ob ¬ ject, in order to find out whether the detected ob ¬ ject is only the tagged object or whether it is a combination of the tagged object and another ob- ject, which might be a human carrying the tagged obj ect .

Judgement can also be made or inferred from a pre ¬ defined computer model, for example, CAD or virtual reality model for the already known devices and fixed setup of equipment in the scenery. The com ¬ puter model, considered as a "digital twin" of the real environment, can be periodically rescanned for an update of environment changes.

If pre-defined criteria do not allow for a certain judgment, the possibility should be provided for an operator to declare whether an object qualifies for extended protection or not.

Further objects of the invention are a robot system comprising a robot, a sensor system for surveilling a vicinity of the robot and a processor unit for controlling the robot based on data from the sensor system using the method as described above, and a computer program product comprising program code means which enable a computer to perform the method . Further features and advantages of the invention will become apparent from the subsequent descrip ¬ tion of embodiments, referring to the appended drawings . Fig. 1 is a schematic plan view of a robot and its vicinity; Fig. 2 is a plan view in which an object qualifying for extended protection is entering the vicinity;

Fig. 3 is a diagram illustrating exemplary changes of the qualification of objects in the vicinity of the robot;

Fig. 4 is another plan view of the robot and its vicinity;

Fig. 5 is another diagram illustrating exemplary changes of the qualification of objects; and

Fig. 6 is a flowchart of a method carried out in a processor unit that controls the robot.

Fig. 1 is a schematic plan view of an industrial robot 1 and its working environment. In the embodi ¬ ment shown here, the robot 1 has a stationary base 2, one articulated arm 3 and an end effector 4 at the free end of the arm, but it should be kept in mind that the method which will be described subse ¬ quently is also applicable to more complex systems in which the robot is mobile and/or in which it comprises more than one arm and/or more than one end effector.

The motion of the robot 1 is controlled by a pro ¬ cessor unit 5, typically a microcomputer, in order to carry out a predefined manufacturing task, for example combining each of a series of first work- pieces 6, supplied by a conveyor 7, with a second workpiece 8, taken from a crate 9. The processor unit 5 may comprise two functionally separate but cooperating processor units: A first processor unit for controlling the robot, and a second processor unit for sensor data evaluation. Both can be within the same electronics rack, but can also be set up apart from each other. Functionally both together make up the processor unit 5.

The processor unit 5 is connected to a spatially resolving sensor 10 which is designed and posi ¬ tioned to monitor the vicinity of the robot 1. In the following the vicinity of the robot 1 and the region in space in which objects can be detected by the sensor 10 will be regarded as identical, assum- ing that the sensor 10 has been set up appropriate ¬ ly to detect any object that might approach the ro ¬ bot 1 and collide with it, no matter from which di ¬ rection. In Fig. 1, the sensor 10 is represented as a single camera. In practice it may comprise sever- al cameras viewing the vicinity of the robot 1 from different directions, so that no object can be hid ¬ den from all cameras at a time by the robot 1, and so that the processor unit 5 can determine coordi ¬ nates of an object seen by more than one camera by triangulation .

Of course any other type of sensor capable of providing spatially resolved information on objects in the vicinity of the robot 1 can be used as or combined into the sensor 10, e.g. a radar device, one or more laser or microwave scanners, weight- sensitive tiles on the floor around base 2, etc.

In Fig. 1 the space in which the robot 1 is working is partially confined by walls or barriers 11. Part of these walls or barriers 11 is in the detection range of sensor 10. One of the walls 11 has a pas- sage 12 through which a moving object might at any time enter the vicinity of the robot 1 and collide with it. While a contact between the object and the robot 1 may be desired and should not be prevented, it should be made sure that when such a contact oc ¬ curs, the robot 1 is at rest and cannot damage the object. Therefore, at all times, the processor unit 5 defines a safety zone 13 which extends from the passage 12 into the vicinity and which it will not allow the robot 1 to enter. The dimension of the safety zone 13 is determined by the condition that whenever an object enters the vicinity through pas ¬ sage 12 with a reasonable speed, e.g. of a quickly walking human, it shall be possible to bring the robot to rest before the object and the robot 1 can touch each other.

The size of the safety zone 13 may vary depending on the instantaneous speed of the robot. The faster the robot 1 moves, the longer is the time needed to bring it to a standstill, and the longer is the distance an intruding object would cover in that time; therefore the width of the safety zone 13 can be adapted to the robot speed.

Further, the direction of the robot movement can be taken into account, making the safety zone 13 larg ¬ er when the robot 1 is approaching the passage 12, and making it narrower when the robot 1 is moving away from it.

The detection range of the sensor 10 can be limited not only by walls 11 but also by objects such as e.g. an electronics rack 14 accommodating processor unit 5, which might be movable but are usually not moved while the robot 1 is operating. In the plan view of Fig. 1 there is a passage 15 between the rack 14 and the conveyor 7 which is hidden from the sensor 10 and from which a further safety zone 16 defined according to the same criteria as described above for safety zone 12 extends into the vicinity of the robot 1.

While it may be possible for the processor unit 5 to determine the position of the walls 11, the rack 14 and the conveyor 7 from data provided by sensor 10, the passages 12, 15 cannot be detected by the sensor 10. Therefore, when the system of Fig. 1 is set up, the processor unit 5 may be provided with a geometrical model of the vicinity in which all pas ¬ sages by which a human might enter are specified.

The system of Fig. 1 has at least one more passage, 17, namely the one by which first workpieces 6 en ¬ ter the vicinity. A safety zone assigned to this passage 17 might be a nuisance for operation of the robot 1, since it might have to be considered when controlling the movement of the robot 1 between the crate 9 and the conveyor 7. Therefore, this passage 17 has no safety zone assigned to it. The lack of a safety zone here may be justified pragmatically by the assumption that no human will voluntarily sit on the conveyor and that therefore the possibility that an object qualifying for extended protection might approach the robot 1 through passage 17 can be neglected, or it may be justified technically by blocking access to an upstream portion of the conveyor 7, beyond passage 17, or by making passage 17 too small for a human to pass.

When the system is operating, the sensor 10 will detect any first workpiece 6 that enters the vicin ¬ ity by passage 17, but as this passage 17 has no safety zone assigned to it, the workpieces 6 are automatically detected as what they are, i.e. as objects which do not have to be protected from con ¬ tact with the robot but are there to be handled by it .

Things are different if an object 18 enters the vi ¬ cinity through a passage having a safety zone, as shown in Fig. 2. Here, the object 18 entering the vicinity by passage 12 is a person 19 carrying a crate 20. While the object 18 is not in full view of the sensor 10, the processor unit 5 cannot be certain of the nature of the object 18 and must ex ¬ pect it to move unpredictably. Therefore, as a mat ¬ ter of precaution, it regards the object 18 as qualified for extended protection and therefore de ¬ fines a safety zone 21 around said object 18 ac ¬ cording to the same criteria as described above for safety zone 13. The object 18 will stay surrounded by this safety zone 21 while it moves in the vicin- ity of the robot 1 and e.g. replaces the old crate 9 by the new one 20, filled with fresh second work- pieces 8.

Fig. 3 illustrates how the judgment of the proces- sor unit 5 on the qualification of the various ob ¬ jects 18, 9 in the vicinity evolves in the course of this replacement. At the left-hand side of the diagram, at the beginning of the process (step a) , there are two symbols, a square one at a long dis- tance d from the robot 1, representative of the ob ¬ ject 18, and a round one, at a small distance, rep ¬ resentative of crate 9. The symbol of object 18 is cross-hatched, indicative of its qualification for extended protection; the symbol of crate 9 is emp- ty, since the crate 9 does not qualify, but rather, the workpieces 8 in it are there to be manipulated by the robot 1. The qualification statuses of object 18 and crate 9 remain unchanged in steps b and c, while the person 19 approaches crate 9. When the distance between both is small enough for the crate 9 to be within an arm's length of the person 19, i.e. if the dis ¬ tance is less than 0.85 m, the processor unit 5 changes the status of the crate 9, indicated by cross-hatching of its symbol in Fig. 3 in step d. Now the safety zone 21 is extended to envelop also crate 9, as shown in Fig. 4. The extended safety zone 21 prevents the robot 1 from fetching a new second workpiece from crate 9. The person 19 puts down the new crate 20, whereby the crate 20 and the person 19 become distinguisha ¬ ble as two objects by the sensor 10 and the proces ¬ sor unit 5. This is represented by distinct symbols for the crate 20 and for the person 19 appearing in step e.

In step f, the person 19 takes up the old crate 9, represented by a new symbol 22 of a combined object formed of the person 19 and the crate 9, replacing that of crate 9. In the following steps, the com ¬ bined object 21, i.e. the person 19 and the old crate 9 leave the vicinity, while the new crate 20 stays behind. If the new crate 19 continued to be protected from contact with the robot 1, the robot 1 couldn't con ¬ tinue work with the new workpieces 8 in crate 20. Therefore, appropriate criteria must be defined which enable the processor unit to decide that an object no longer qualifies for extended protection and to reduce its safety zone. One possible criterion for such a decision is the absence of motion. If no motion of an object is de ¬ tected within a predetermined time span, the object can be assumed to be inanimate and therefore not to need a degree of protection which would prevent a contact with the robot 1 even if the object sudden ¬ ly started to move. The duration of such a time span depends on the precision with which the sensor 10 is capable of detecting motion. If the sensor 10 is sensitive enough to detect micro-motion related to heartbeat or respiration, a duration of some seconds may be sufficient for a reliable judgment.

Judgment might also be based on temperature. If the object to be judged is or comprises a human, then at least part of its surface should emanate infra ¬ red radiation. So, if the sensor 10 comprises a py ¬ rometer, and the pyrometer detects no part of the surface of the object with a temperature that might fit a human body, then the object is most likely inanimate and can be judged not to qualify for ex ¬ tended protection.

Another criterion which is easy to evaluate is the weight of the object; if it is noticeably different from that of a human, the object can be judged not to qualify. A disadvantage of this criterion is that the weight cannot be sensed remotely, so that the vicinity of the robot must be equipped at con- siderable cost with appropriate sensors, such as the weight-sensitive tiles mentioned above.

If the sensor 10 comprises a camera, image pro ¬ cessing techniques can be used to extract infor- mation on the shape and/or dimensions of the object from the data provided by the camera, and extended protection may be ruled out if these data allow to exclude the presence of a human, e.g. in case of the crate 19, because it is too small for a human to fit in. There might be the case that an object which a per ¬ son brings into the vicinity of the robot 1 moves even after the person has left it behind, e.g. a fan. If motion is a criterion which will prevent the processor unit 5 from changing the status of an object to non-qualifying, the fan will always enjoy a wide safety zone around it which may impose se ¬ vere limits on the mobility of the robot 1 and de ¬ crease its productivity. Therefore, the processor unit 5 can have a user interface 23 where a user can inspect the objects which the processor unit 5 has found to qualify for extended protection, e.g. by having them highlighted in an image of the vi ¬ cinity displayed on a screen, and, if necessary, change the status of an object.

According to a preferred embodiment, the crates 9, 20 carry tags 24 that can be read by the sensor 10 and that specify the nature of the object to which they are attached and/or some of its characteris- tics. Such a tag can be an optical tag, e.g. a QR tag, which can be read using the camera of sensor 10. Preferably, it is a tag which communicates by radio, such as an RFID tag, a Bluetooth or NFC device, and the processor unit 5 has a radio inter- face 25 connected to it for communicating with the tags 24. In that case, when the tag 24 enters the range of radio interface 25, the processor unit 5 receives data from the tag 24, from which it can draw conclusions on characteristics detectable by sensor 10, of an object which may be about to enter or may already have entered the vicinity of the ro ¬ bot 1. These data may e.g. specify that the object carrying tag 24 is a crate, its size, and other characteristics. When the object 18, comprising the crate 20 and the person 19 carrying it, enters the vicinity, and the sensor 10 detects the character- istics of this object 18, the processor unit 5 finds that these do not match object 18. Therefore, while the person 19 is moving in the vicinity of the robot 1 carrying the crate, data from the tag 24 will never induce the processor unit 5 to deny extended protection. However, as soon as the crate 20 is put down and becomes detectable as an object of its own, the processor unit 5 will find the data from tag 24 to match this object, so that the crate 20 is found not to qualify for extended protection as soon as the distance between it and the person 19 is long enough, in step k of Fig. 3.

An object which does not qualify for extended pro ¬ tection can still have a safety zone, but the width of the safety zone can be substantially reduced; in case of the crate 20 such a safety zone may prevent the robot from touching the walls of the crate 19 while being small enough to allow the robot to reach in the crate 20 and take out workpieces 8 without violating the safety zone. Alternatively, a non-qualifying object may have no safety zone at all; in that case the robot 1 would be allowed to touch and handle it, e.g. to grab the crate 20 and to pull it from where the person 19 put it to a lo- cation closer to the conveyor 7.

The diagram of Fig. 5 relates to a situation in which an object moves through the vicinity of the robot 1 without actually interacting physically with other objects therein, e.g. in which person 19 comes in through passage 12, works on the robot 1 and walks out again. The first steps of the diagram are similar to those of Fig. 3; on his way to the end effector 4 of robot 1, the person 19 passes within an arm' s length of crate 9 and thereby makes it qualify for extended protection, too, in step d. However, since the person 19 leaves the vicinity again without touching the crate 9, the status of the crate 9 is reversed to not qualifying in step g. This can be done based on one or a combination of the criteria described above, so that after the person 19 has left, the robot 1 returns to normal operation after a certain delay. Alternatively, if the qualifying status has been conferred to the crate 9 by the person 19 merely coming close, and if no contact between these person and the crate has occurred, so that the two objects have never become undistinguishable, the status of the crate 9 can be returned no non-qualifying as soon as the person 19 starts to move away from it or, at lat ¬ est, when the distance between the two objects 9, 19 increases above said arm's length or some other appropriately defined threshold, thus minimizing the delay with which normal operation is resumed.

Fig. 6 is a flowchart of a control method of the processor unit 5 which is continuously re-iterated in order to achieve the above-described behaviour.

In step SI, the processor unit 5 fetches current data on the vicinity of the robot, e.g. an image or, if the sensor 10 comprises multiple cameras, a set of images. Based on a pre-defined geometrical model of the vicinity, it identifies the permanent objects which form this vicinity, such as the walls 11, the conveyor 7, the rack 14, and possibly also the robot 1 itself (S2) . Any image details not ac ¬ counted for by these objects must be related to non-permanent objects. If any of these objects was already present in a previous iteration of the method, data relating to it will have been recorded by the processor unit 5; if a non-permanent object exists which has moved since the previous iteration or which has newly entered the vicinity, it will be detected in step S3 since its position as derived from the current sensor data will not match any of the recorded position data. If such an object exists and its position is found in step S4 to be closer to a passage 12, 15 or 17 than to any recently recorded position, the object is judged to be new in the vicinity. If the passage was one of the passages 12, 15 that can be passed by a human (S5) , the object is possibly human and is therefore judged to qualify for extended protec ¬ tion in step S6; else, if the entry point was pas ¬ sage 17, it is judged not to qualify in step S7. The entry point is included in the object data rec- orded in step S8.

If in step S4 the object is found to be closer to a previously existing object than to a passage, then the object is regarded as being identical to or be- ing a fragment of said previously existing object, and the protection status and the entry point of said previously existing object are copied to the present object (S9) . In this way, while the object 18 of Fig. 2 moves through the vicinity, it is al- ways judged as qualifying for extended protection whereas the workpieces 6 moving on conveyor 7 do not. Due to the possibility of fragmentation, when the person 19 sets down crate 20 in the scenario of Fig. 4 and moves away from it, both become objects of their own and are judged to qualify for extended protection . Step S10 checks whether the copied status of the object is not qualifying and its entry point is one of passages 12, 15. If both conditions are met, a check is carried out in step Sll whether within an arm's length of the there is another object which does qualify, and if exists, the status of the ob ¬ ject is changed to qualifying (12) . This is what happened to crate 9 in step d of Fig. 3. If a current data set is found in step S3 to be identical to a previously recorded one, it must be ¬ long to an object which hasn't moved since the pre ¬ vious iteration. If the status of the object is not qualifying (S13) , nothing happens to it. Else the processing unit 5 examines (S14), using one or more of the criteria described above, whether the status may be changed, and changes it to non-qualifying (S15) if the criteria are met. Step S16 checks whether there is an object whose data have not yet been processed, and if there is, the method returns to step S3 to process another set of data; if there isn't, the iteration is fin ¬ ished .