Robotic Arm

Abstract

In general terms, the present invention provides a passively compliant robotic arm having one or more variable stiffness joints controllable by first and second bi-directional actuators that can be independently operated. Each bi-directional actuator may be operable in a first configuration to urge the joint in a first direction, and in a second configuration to urge the joint in a second direction opposite to the first direction. The bi-directional actuators may be operated in a cooperating mode (high torque mode) in which they work in tandem (i.e. both in the first configuration or second configuration) to double the available torque output. The bi-directional actuators may also (or alternatively) be operated in a high stiffness mode (antagonist mode) in which they counter-act each other by operating so that they oppose one another (i.e. one in the first configuration and the other in the second configuration). The high torque mode may be utilised for an initial portion of a movement trajectory, and the antagonist mode for a final portion of the movement trajectory. The relatively high stiffness in the high stiffness/antagonist mode results from the combined effects of the non-linear force-deflection relationship of the first and second resilient members. The resilient members may each comprise an elastic element, tendon or other resilient member that can be stretched (elongated) to increase tension therein and thereby urge the joint to move.

Claims

1. A robotic arm comprising: a joint permitting movement between a first link and a second link; and first and second bi-directional actuators, each bi-directional actuator comprising a first resilient member actuatable by an increase in tension to urge the joint to move in a first direction and a second resilient member actuatable by an increase in tension to urge the joint to move in a second direction opposite to the first direction, each of the bi-directional actuators being controllable to increase tension in one of the first and second resilient members while decreasing tension in the other of the first and second resilient members, wherein the first resilient members and second resilient members have a monotonically increasing non-linear relationship between applied force and resulting elongation, and comprise a composite material having a generally elastic portion and a relatively stiff portion.

2. A robotic arm according to claim 1, wherein the first and second bi-directional actuators are controllable to operate in a high torque mode in which the first and second bi-directional actuators each provide tension in their respective first resilient members.

3. A robotic arm according to claim 1, wherein the first and second bi-directional actuators are controllable to operate in an antagonist mode in which the first bi-directional actuator provides tension in its first resilient member while the second bi-directional actuator provides tension in its second resilient member.

4. A robotic arm according to claim 1, wherein the relatively stiff portion provides an increase in stiffness of the respective resilient member with elongation of the resilient member.

5. A robotic arm according to claim 1, wherein the elastic portion comprises a core of the composite material and the relatively stiff portion comprises an outer surrounding portion.

6. A robotic arm according to claim 1, wherein the relatively stiff portion comprises a spiral of material encasing the elastic portion.

7. A robotic arm according to claim 1, wherein the elastic portion comprises an elastomer, such as a thermoplastic elastomer.

8. A robotic arm according to claim 1, wherein the relatively stiff portion comprises a polymer, such as a thermoplastic polymer.

9. A robotic arm according to claim 1, wherein the joint permits the second link to pivot relative to the first link about a pivot axis of the joint located between a first anchor point and a second anchor point of the second link, wherein in each of the first and second bi-directional actuators the first resilient member extends between the first anchor point and the first link and the second resilient member extends between the second anchor point and the first link.

10. A robotic arm according to claim 1, further comprising: an end effector arranged to engage an object, movement of the joint causing movement of the end effector; a first sensing apparatus arranged to sense an initial estimated location of the object; and a second sensing apparatus arranged to sense a final sensed location of the object, the second sensor being arranged to move in tandem with the end effector, wherein the first and second bi-directional actuators are arranged to move the end effector initially to the initial estimated location, preferably in the high torque mode, and finally to the final sensed location, preferably in the antagonist mode.

11. A robotic arm according to claim 3, further comprising an end effector arranged to engage an object, movement of the joint causing movement of the end effector, wherein the first and second bi-directional actuators are controllable to move the end effector towards an object to be engaged initially in the high torque mode, and finally in the antagonist mode.

12. A method of controlling a robotic arm comprising a joint permitting movement between a first link and a second link, and first and second bi-directional actuators, each of the first and second bi-directional actuators comprising a first resilient member actuatable by an increase in tension to urge the joint to move in a first direction and a second resilient member actuatable by an increase in tension to urge the joint to move in a second direction opposite to the first direction, wherein the first resilient members and second resilient members have a non-linear relationship between applied force and resulting elongation, and comprise a composite material having a generally elastic portion and a relatively stiff portion, the method including the steps of: controlling the first bi-directional actuator to increase tension in one of the first and second resilient members thereof while decreasing tension in the other of the first and second resilient members; and controlling the second bi-directional actuator to increase tension in one of the first and second resilient members thereof while decreasing tension in the other of the first and second resilient members.

13. A method according to claim 12, including the step of controlling the first and second bi-directional actuators to operate in a high torque mode by increasing tension in their respective first resilient members and decreasing tension in their respective second resilient members.

14. A method according to claim 12, including the step of controlling the first and second bi-directional actuators in an antagonist mode by, in the first bi-directional actuator, increasing tension in the first resilient member and decreasing tension in the second resilient member, while simultaneously, in the second bi-directional actuator, increasing tension in the second resilient member and decreasing tension in the first resilient member.

15. A method according to claim 12, wherein the robotic arm comprises a robotic arm according to claim 1.

16. A method according to claim 12, comprising controlling the robotic arm to move an end effector of the robotic arm to an object, the method including the further steps of: a) sensing an initial estimated location of the object using a first sensing apparatus; b) moving the at least one joint to move the end effector to a nearby location in the vicinity of the initial estimated location, preferably by controlling the first and second bi-directional actuators to operate in the high torque mode; c) sensing a final sensed location of the object using a second sensing apparatus configured to move in tandem with the end effector; and d) moving the at least one joint to move the end effector to the final sensed location, preferably by controlling the first and second bi-directional actuators in the antagonist mode.

17. A method according to claim 12, comprising controlling the robotic arm to move an end effector of the robotic arm to an object, the method including controlling the first and second bi-directional actuators to move the end effector towards the object initially in the high torque mode, and finally in the antagonist mode.

18. A system for picking fruit or vegetables, comprising a moveable base supporting one or more robotic arms according to claim 1.

19-54. (canceled)

55. A method of picking fruit or vegetables comprising a method of controlling a robotic arm according to claim 16, wherein the object comprises a fruit or vegetable.

56. A method of picking fruit or vegetables comprising a method of controlling a robotic arm according to claim 17, wherein the object comprises a fruit or vegetable.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0118] One or more embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

[0119] FIG. 1 shows a side view of a robotic arm according to an embodiment of the invention;

[0120] FIG. 2 shows an isometric view of the robotic arm of FIG. 1;

[0121] FIG. 3 shows an embodiment of a resilient member suitable for use in embodiments of the invention;

[0122] FIGS. 4A and 4B schematically illustrate a trajectory carried out by an end effector of a robotic arm according to an embodiment of the invention, and a change in stiffness of one or more joints within that arm during movement through the trajectory, respectively;

[0123] FIGS. 5A and 5B illustrate embodiments of control architecture for a ballistic phase (FIG. 5A) and a closed-loop joint control phase (FIG. 5B) of movement of one or more joints of a robotic arm according to an embodiment of the invention; and

[0124] FIGS. 6A, 6B and 6C illustrate a fruit-picking system incorporating multiple robotic arms according to an embodiment of the invention.

DETAILED DESCRIPTION

[0125] FIGS. 1 and 2 illustrate side and isometric views, respectively, of a robotic arm 100 according to an embodiment of the invention. The arm comprises a base 10 via which the arm 100 can be mounted to a structure, and via which the arm can receive an electricity supply (not shown).

[0126] The arm 100 has a plurality of articulated joints, including a shoulder joint 20 and an elbow joint 30. Each joint provides movement with six degrees of freedom; in Cartesian space this corresponds to displacement along x, y and z axes, and rotation about each of the x, y and z axes. Movement of the joints controls the position and orientation of the end effector 50, which in this embodiment comprises two opposing rigid finger members 52 with a pliable pad 54 at a free end thereof. The finger members 52 move in a pincer configuration to grip an object (not shown) between the pliable pads 54.

[0127] The shoulder joint 20 and elbow joint 30 each comprise a variable stiffness joint which enables the passive compliance of the joint (i.e. the resistance to deflection resulting from an externally applied force/torque) to be controlled. The principles of the variable stiffness joint will be described below in relation to the elbow joint 30, though the skilled reader will readily understand how these principles may be applied to the shoulder joint 20 or to any other joint in the robotic arm 100.

[0128] The elbow joint 30 is controllable to provide relative movement between a rigid first link 32 that extends towards the shoulder joint 20, and a rigid second link 34 that extends towards the end effector 50. The second link 34 pivots relative to the first link 32 about a pivot axis 33.

[0129] Relative movement between the first and second links is controlled by first 40a and second 40b bi-directional actuators that together enable the elbow joint 30 to be a variable stiffness joint. Each bi-directional actuator 40a, 40b comprises a motor (not visible in FIGS. 1 and 2) driving a pulley wheel 42a, 42b mounted on the first link 32, each of which engages a flexible actuation link 44a, 44b. The actuation links 44a, 44b each comprise an elongate cord that is tethered at one end to a first anchor point 36a, 36b on the second link 34 and at the other end to a second anchor point 38a, 38b on the second link 34, the first and second anchor points being on either side of the pivot axis 33. Each actuation link has a generally non-extensible portion 45a, 45b that engages the pulley wheel, a first tendon portion 46a, 46b (also referred to as a first resilient member 46a, 46b) that extends between the pulley wheel and the first anchor point, and a second tendon portion 48a, 48b (also referred to as a second resilient member 48a, 48b) that extends between the pulley wheel and the second anchor point.

[0130] The first 46a, 46b and second 48a, 48b tendon portions each have a monotonically increasing non-linear relationship between applied force and resulting elongation. That is, the resistance to elongation (stiffness) increases with increased applied force.

[0131] An embodiment of the first 46a, 46b and second 48a, 48b tendon portions is illustrated in FIG. 3. Each tendon portion comprises an elongate elastic core 47 around which a helical- or spiral-shaped stiffening portion 49 is wrapped. The elastic core 47 has a circular cross-section and is formed from an elastomeric material that is able to provide a significant degree of elongation (e.g. up to 700% increase in length) when under tension and return to its original shape and size when the tension is removed. An appropriate material for the elastic core 47 is a TPE thermoplastic elastomer such as Filaflex, produced by Recreus. In contrast, the material from which the stiffening portion 49 is made is generally non-elastic; a suitable material is nylon.

[0132] When an elongating tensile force is applied to each of the tendons, the spiral shape of the stiffening portion 49 means that it becomes longer in the longitudinal direction (the direction in which the force is applied) while becoming narrower in the lateral direction (perpendicular to the longitudinal direction). This lateral contraction is resisted by the elastic core 47, and this resistance and the consequential deformation of the elastic portion causes a progressive increase in resistance to elongation with an increase in applied force. In this way, as the tendon is elongated, the elastic portion initially carries the majority of the load, but as the elongation increases the relatively stiff portion carries a progressively higher proportion of the load to provide an increasing resistance to further elongation.

[0133] In use, each bi-directional actuator 40a, 40b is able to provide movement of the elbow joint 30 in both a first direction (clockwise movement of the second link 34 relative to the first link 32, as seen in FIGS. 1 and 2) and a second direction opposite to the first direction. Movement in the first direction is caused by operating the motor to turn the pulley 42a, 42b (in the clockwise direction) to increase tension in the first tendon 46a, 46b and thereby shorten the distance between the first anchor point 36a, 36b and the pulley, while lengthening the distance between the second anchor point 38a, 38b and the pulley. Similarly, movement in the second direction is caused by operating the motor to turn the pulley 42a, 42b (in an anti-clockwise direction) to increase tension in the second tendon 48a, 48b and thereby shorten the distance between the second anchor point 38a, 38b and the pulley, while lengthening the distance between the first anchor point 36a, 36b and the pulley

[0134] In this way, each of the bi-directional actuators 40a, 40b is operable in a first configuration to urge the elbow joint 30 in the first direction, and in a second configuration to urge the joint in the second direction. By controlling movement of the joint via independent control of both the first 40a and second 40b bi-directional actuators it is possible to vary the stiffness (the resistance to externally-applied forces) of the joint while controlling its position.

[0135] That is, in a high torque mode (cooperating mode) each of the bi-directional actuators 40a, 40b may be operated to work in tandem (i.e. both in the first configuration or second configuration) to maximise the available torque output. In this mode the joint has a relatively low stiffness and relatively high passive compliance. At the other end of the spectrum, the bi-directional actuators 40a, 40b may be operated in a high stiffness mode (antagonist mode) in which they counter-act each other (i.e. one in the first configuration and the other in the second configuration). In this mode the joint has a relatively high stiffness and relatively low passive compliance.

[0136] The bi-directional actuators 40a, 40b can also be operated at any point along the continuous spectrum between the high torque mode and high stiffness mode, so that as the stiffness of the joint is reduced the available torque output can be increased, and vice versa. In this way, when low joint stiffness is required (such as during ballistic phase movements described below) the torque output of the joint can be maximised.

[0137] The relationship between the first 40a and second 40b bi-directional actuators can be described by way of the differential position, p, of the first 42a and second 42b pulleys. That is, in the high torque mode the differential position may have a maximum value of 1, in the high stiffness mode a maximum value of 1, and values between 1 and -1 may represent differential positions in the spectrum between these extremes.

[0138] In alternative embodiments, each of the bi-directional actuators 40a, 40b may be replaced by a uni-directional actuator (not illustrated). For example, the first uni-directional actuator 40a may comprise only the first tendon 46a and no second tendon, and the second uni-directional actuator 40b may comprise only the second tendon 48b and no first tendon. In this way, movement of the elbow joint 30 in the first direction may be achieved by operating the pulley 42a of the first uni-directional actuator 40a to increase tension in the first tendon 46a, and movement of the elbow joint 30 in the second direction may be achieved by operating the pulley 42b of the second uni-directional actuator 40b to increase tension in the second tendon 48b. Moreover, the first and second uni-directional actuators may be operated together to control the overall stiffness of the elbow joint 30 in a similar manner to that described above in relation to the high stiffness mode (antagonist mode) of the bi-directional actuator embodiment.

[0139] Each of the joints of the arm 100, including the shoulder joint 20, may have a variable stiffness joint as described above in relation to the elbow joint 30. The elbow joint 30 is described merely as being exemplary of any variable stiffness joint in the arm 100.

[0140] The robotic arm 100 also comprises a sensor-control phase stereo camera 60 and colour camera 70 mounted on an end link of the arm so that they move in tandem with the end effector. In this way, the sensor-control phase stereo camera 60 and colour camera provide continuous images of the region between the pliable pads 54 of the end effector 50, and a limited portion of the environment surrounding the end effector 50. The sensor-control stereo camera 60 and colour camera 70 are used in the sensor control (final approach) phase of movement of the robotic arm 100, as described further below.

[0141] Each robotic arm 100 also has an associated joint control phase stereo camera (250 in FIGS. 6A-C; not shown in FIGS. 1 and 2) that is positioned in a fixed position with respect to the base 10, and provides images including all points accessible by the end effector 50 of that arm 100.

[0142] In use, the robotic arm 100 is controlled to control the position of the end effector 50 by controlling the position of each of the joints, including the elbow joint 30 and shoulder joint 20, while simultaneously controlling the stiffness of each of those joints.

[0143] FIGS. 4A and 4B illustrate an embodiment of the invention in which the robotic arm 100 is controlled via a four-phase movement. This four-phase movement is considered to be particularly suitable for applications in which an object is to be grasped by the end effector 50, such as fruit-picking applications.

[0144] FIG. 4A illustrates an example trajectory for a fruit-picking movement, while FIG. 4B schematically illustrates the change in joint stiffness at the elbow joint 30 (or shoulder joint 20 or other joint) over that trajectory. The end effector 50 starts at t.sub.0 and travels in a ballistic phase through t.sub.1 and t.sub.2 to t.sub.3. The trajectory from t.sub.3 to t.sub.4 represents a closed-loop joint control phase, while the trajectory from t.sub.4 to t.sub.6 represents a sensor-control phase. At t.sub.6 the end effector 50 grasps the fruit, and the trajectory from t.sub.6 to t.sub.7 represents the detachment phase where the fruit is detached from the stem, cane, bush, vine, stem or tree on which it has grown.

[0145] The control architecture for the ballistic phase and closed-loop joint control phase are illustrated in FIGS. 5A and 5B, respectively.

[0146] The ballistic phase (FIG. 5A) has as inputs a desired joint angle, .sub.d, and desired joint stiffness, c. An inverse joint model is used to map the desired joint angle and desired joint stiffness to the corresponding differential position, p, of the pulleys 24a, 42b, based on the desired joint angle and desired joint stiffness. Then, an equilibrium equation is used to determine the angular position, .sub.1, of the pulley 42a of the first bi-directional actuator 40a and the angular position, .sub.2, of the pulley 42b of the second bi-directional actuator 40b that will achieve both the differential position, p, and the desired joint stiffness, c.

[0147] The output of the ballistic phase of joint control at t.sub.3 is a joint angle, , which is close to .sub.d, preferably more than 50% of .sub.d, and ideally 60%, 70%, 80% or 85% or more of .sub.d. The ballistic phase thus moves the end effector 50 to a nearby location in the vicinity of the initial estimated location of the fruit to be picked, as described further below.

[0148] The closed-loop joint control phase (FIG. 5B) also has as inputs the desired joint angle, .sub.d, and desired joint stiffness, c. The current joint angle, , is fed back to the controller to determine a joint angle difference, .sub., between the current joint angle, , and the desired joint angle, .sub.d, and thereby reduce that difference, A. A feedback control law step determines, based on the joint angle difference, .sub., a change in differential position, p, that will reduce the joint angle difference .sub.. This is then transformed into a differential position, p, which is used by an equilibrium equation to determine the angular position, .sub.1, of the pulley 42a of the first bi-directional actuator 40a and the angular position, .sub.2, of the pulley 42b of the second bi-directional actuator 40b that will achieve both the differential position, p, and the desired joint stiffness, c.

[0149] The output of the closed-loop joint control phase of joint control at t.sub.4 is a joint angle, , that is even closer to .sub.d, preferably exactly at .sub.d. However, since the joints of the arm 100 are not rigid, but are compliant to a greater or lesser degree, achieving the desired joint angle, .sub.d, may not result in the end effector being located in precisely in the right position to grasp the fruit. Moreover, the target may be moving (e.g. by the action of wind) and/or the location data provided by the joint control phase stereo camera may be inaccurate. The sensor-control phase corrects for these errors at the end of the trajectory, from t.sub.4 to t.sub.6.

[0150] Movement through the trajectory of FIG. 4A will now be described, by way of an example of how the arm is controlled in use.

[0151] At t.sub.0 the ballistic phase stereo camera is used to generate a three-dimensional point cloud containing target positions of fruit to be picked (or other target positions in other applications). Each target has a position in a Cartesian coordinate system with an origin, or reference point, at or near the base 10 of the arm 100. Once a point of interest has been selected from the point cloud, the ballistic phase trajectory to t.sub.3 is generated and the angular positions, .sub.1, of the pulley 42a of the first bi-directional actuator 40a and the angular position, .sub.2, of the pulley 42b of the second bi-directional actuator 40b at each of t.sub.1, t.sub.2 and t.sub.3 are calculated using the control architecture described above in relation to FIG. 5A.

[0152] The joint is then moved from t.sub.1 to t.sub.2 and then to t.sub.3 by controlling the bi-directional actuators 40a, 40b to reach the calculated angular positions and thereby ensure the relative positions of the pulleys 42a, 42b at each point of the ballistic phase trajectory is such that the joint has the desired joint pose with the desired level of stiffness.

[0153] It can be seen from FIG. 4B that the stiffness of the joint is relatively low during the ballistic phase, though rises from t.sub.3 to t.sub.4 on the approach to the closed-loop joint control phase. During this phase the joint moves relatively fast and the open loop control further speeds up arrival at t.sub.3 by reducing the number of control steps required. This fast movement with limited control has the potential to result in collisions between the arm 100 and an external body, such as a person or structure. However, the relatively low joint stiffness ensures that the arm 100 has a relatively high level of passive compliance during the ballistic phase, with the result that such collisions should not cause damage to either the arm 100 or the external body. The high torque mode may be utilised during the ballistic phase to maximise the torque available to move the joint.

[0154] At t.sub.3 the closed loop joint control phase commences. The control architecture described above in relation to FIG. 5B is used to calculate each change in angular position, .sub.1, of the pulley 42a and angular position, .sub.2, of the pulley 42b required to reduce the joint angle difference, . The joint is progressively moved towards t.sub.4 by controlling the bi-directional actuators 40a, 40b to reach the calculated angular positions, and repeating until the joint angle difference, .sub., is zero or within an allowable margin of zero.

[0155] It can be seen from FIG. 4B that the stiffness of the joint continues to rise in the closed-loop joint control phase from t.sub.4 to t.sub.5, to a level at which the joint stiffness is relatively high. Passive compliance in this high stiffness mode is thus reduced, but accuracy of joint position increases.

[0156] In the sensor-control phase from t.sub.4 to t.sub.6 the high stiffness mode is maintained. During this phase the movement of the joint is controlled based on visual data obtained by the sensor-control phase stereo camera 60 and colour camera 70. Images obtained by the colour camera 70 are analysed to identify the fruit to be picked using image recognition algorithms. For example, a cluster of pixels in a certain colour range or in a certain pattern may indicate the presence of a fruit. The image may also be analysed to determine whether the identified fruit is ripe and/or whether it is blemished, and therefore whether or not it should be picked.

[0157] Once the fruit has been identified images obtained by the approach phase stereo camera 60 are analysed to determine the sensed location of the identified fruit in a Cartesian coordinate system that is local to the end effector.

[0158] The determined sensed location is compared to the known location of the end effector 50, and the trajectory to be traveled by the end effector 50 is calculated. The angular positions of the pulleys 42a, 42b required to achieve the joint positions required to achieve movement along the calculated trajectory are determined, and the joint is moved to move the end effector to the sensed location. The sensed location may change over time as new data from the cameras 60, 70 is obtained. For example, the fruit may be moving slightly, or the accuracy of the determined location may improve as the end effector gets closer to the fruit. This process may therefore be repeated until the end effector 50 reaches a final location in which it is able to grasp the fruit.

[0159] By using sensors (stereo camera 60 and colour camera 70) that are located in a fixed position relative to the end effector (i.e. able to move in tandem with the end effector) it is possible to calculate the trajectory to be traveled by the end effector in the sensor-control phase within a local coordinate system, which reduces the processing steps required to calculate the necessary joint movements and thereby maximises the speed of motion of the end effector in the sensor-control phase.

[0160] Moreover, during the sensor-control phase the position (angle) of the joint may be controlled by open loop control or by closed loop control. Open loop control is preferred, since this will result in fewer control commands, quicker processing, and thus quicker movement of the joint.

[0161] In the detachment phase from t.sub.6 to t.sub.7 the end effector 50 is rapidly moved downwardly to detach the fruit. Joint movements in this phase are controlled in a similar way to the ballistic phase, via open loop control. It can be seen from FIG. 4B that the joint stiffness is rapidly decreased to a minimum stiffness level in this phase. The rapid decrease in joint stiffness may be caused by a rapid release of highly pre-tensioned tendons in the bi-directional actuators 40a, 40b. This rapid release provides an explosive release of energy at the initial part of the detachment phase, which may help to detach the fruit from its stem.

[0162] FIGS. 6A, 6B and 6C illustrate an embodiment of a fruit- or vegetable-picking system according to the invention. The system comprises a multi-arm mobile platform 200 which is movable by way of wheels 210 or alternatively by way of a rail or gantry system (not shown). The platform 200 supports four robotic arms 100 according to the first embodiment described above stacked vertically above one another, and each mounted via their base 10 to a vertical support 220 of the platform 200. Each robotic arm 100 delivers picked fruit (or vegetables) to a dedicated fruit storage container 230, such as a punnet or tray. The platform 200 also supports a cooling storage unit 240 in which the fruit storage containers 230 are placed once full, in order to maximise the shelf life of the fruit.

[0163] Each robotic arm 100 has an associated ballistic phase stereo camera 250 that provides images including all points accessible by the end effector 50 of that arm 100. Each arm 100 also includes an LED light source (not shown) mounted on the arm so as to have a fixed position relative to the end effector 50, and so as to illuminate an area encompassed by images captured by the colour camera 70 and approach phase stereo camera 60. This illumination enables fruit or vegetable picking in dark conditions, such as during the night, and also helps to control the light conditions to prevent fluctuations in data quality due to variations in the environmental light quality.