Related Applications

This application claims the benefit of the filing date of U.S. Application No.

61/931 ,243, filed January 24, 2014, and U.S. Application No. 62/033,904, filed August 6, 2014, the contents of which are incorporated herein by reference in their entirety.

Field

This invention relates to control of excavation/loading vehicles. In particular, this invention relates to autonomous or semi-autonomous control of excavation/loading vehicles.

Background

Autonomous (or robotic) excavation/loading vehicles are of interest in the mining and construction industries, where the aim is to remove operators from hazardous environments, improve machine utilization and productivity, and reduce maintenance costs. Autonomous excavation is also of interest in lunar or planetary exploration, where excavation cannot easily be carried out by remote control.

In mining and construction, autonomous excavation commonly involves excavation in fragmented rock using a load-haul-dump (LHD) machine. What makes robotic excavation challenging is the nature of the bucket-rock interactions. Performance is strongly influenced by the conditions of interaction between the machine and its environment. For example, the resistance faced by a bucket as it attempts to penetrate a rock pile may vary significantly depending upon the properties of the media (e.g., density and hardness), the rock pile geometry, and the distribution of rock particle sizes and shapes. Indeed, it would be very difficult to predetermine the exact nature of future bucket-rock interactions prior to the execution of any particular excavation operation.

Previous work attempted to automate excavation by using a controller and forces sensed at the bucket to shift between discrete dig paths to excavate a rock pile, or using a compliance controller to dig through soil targets. Both of these controllers were only tested in relatively homogeneous materials, and they did not perform well when sub-surface irregularities were encountered. Hence they were poorly suited for autonomous excavation of typical rock piles, and in real world situations where subsurface obstacles are frequently encountered. Other work using a fuzzy logic behaviour-based controller produced inconsistent results, and is difficult to implement and support as a commercial product. An admittance-based controller using sensed forces to regulate the velocity of the bucket actuator has been proposed, but was never implemented or tested.

Summary

Described herein is a dig controller for an autonomous or semi-autonomous loading vehicle (ALV), comprising: at least one controller that controls a bucket and/or the ALV in accordance with at least one sensor signal, wherein the at least one sensor signal is representative of interaction between the bucket and the rock pile during a dig; wherein the at least one sensor signal is obtained from at least one sensor associated with one or more actuators other than a bucket actuator, or one or more actuated elements. In one embodiment, the dig controller further comprises at least one iterative learning controller (ILC) that uses feedback from at least one previous dig to modify the at least one sensor signal provided to the at least one controller.

Also described herein is a dig controller for an ALV, comprising: at least one controller that controls a bucket and/or the ALV in accordance with at least one sensor signal, wherein the at least one sensor signal is representative of interaction between the bucket and the rock pile during a dig; and at least one ILC that uses feedback from at least one previous dig to modify the at least one sensor signal provided to the at least one controller.

In one embodiment, the at least one controller comprises at least one admittance controller. In one embodiment, the at least one sensor signal is obtained by measuring a force received by a boom actuator. In another embodiment, the at least one sensor signal is obtained by measuring a force received by an actuated element.

In one embodiment, an admittance controller may control velocity of the bucket. At least one admittance controller may comprise an adaptive admittance controller. An adaptive admittance controller may dynamically adjust at least one parameter in response to a difference between a sensor signal and a desired signal. In one embodiment, the at least one controller maps one or more sensor signals to a range of possible bucket velocities or ALV velocities by using at least one of proportional, integral, and derivative control.

In one embodiment, the at least one controller may map a total force error to a range of possible sensed forces using at least one of proportional, integral, and derivative control. In another embodiment, the at least one ILC may map a signal from a previous dig to changes in dig controller response using at least one of proportional, integral, and derivative control.

In one embodiment, the dig controller may modify the at least one sensor signal provided to the controller such that bucket velocity or ALV velocity are changed.

In one embodiment, the dig controller may comprise at least one position controller that controls at least one of movement of the bucket of the ALV to at least one selected pose, and movement of the ALV relative to the rock pile. In another embodiment, the dig controller may comprise a first velocity ILC that perturbs an ALV velocity based on a sensor signal representative of interaction between the bucket and the rock pile during at least one previous dig, and a second ILC that modifies a sensor signal derived from boom and bucket force error measurement of at least one previous dig.

In another embodiment, the dig controller may comprise a first ILC that modifies a sensor signal being provided to a boom admittance controller, and a second ILC that modifies a sensor signal being provided to a bucket admittance controller, wherein modifying is based on feedback from at least one previous dig.

Also described herein is programmed media for use with an ALV dig controller comprising a computer, the programmed media comprising: a computer program stored on non-transitory storage media compatible with the computer, the computer program containing instructions to direct the computer to perform one or more of: receive at least one sensor signal from at least one sensor associated with one or more actuators other than a bucket actuator, or one or more actuated elements; wherein the at least one sensor signal is representative of interaction between the bucket and the rock pile during a dig; and control the bucket and/or the ALV in accordance with the at least one sensor signal.

Also described herein is programmed media for use with an ALV dig controller comprising a computer, the programmed media comprising: a computer program stored on non-transitory storage media compatible with the computer, the computer program containing instructions to direct the computer to perform one or more of: control a bucket and/or the ALV in accordance with at least one sensor signal, wherein the at least one sensor signal is representative of interaction between the bucket and the rock pile during a dig; and direct an ILC to use feedback from at least one previous dig to modify the at least one sensor signal; wherein modifying the at least one sensor signal changes the control of the bucket and/or the ALV.

Also described herein is a method of controlling an ALV, comprising: obtaining at least one sensor signal from at least one sensor associated with one or more actuators other than a bucket actuator, or one or more actuated elements; wherein the at least one sensor signal is representative of interaction between the bucket and the rock pile during a dig; and controlling the bucket and/or the ALV in accordance with the at least one sensor signal.

Also described herein is a method of controlling an ALV, comprising: controlling a bucket and/or the ALV in accordance with at least one sensor signal, wherein the at least one sensor signal is representative of interaction between the bucket and the rock pile during a dig; and modifying the at least one sensor signal using at least one ILC that incorporates feedback from at least one previous dig; wherein modifying the at least one sensor signal changes the control of the bucket and/or the ALV.

In one embodiment the method may include modifying the at least one sensor signal provided to the controller such that bucket velocity or ALV velocity are changed.

Controlling may further comprise dynamically adjusting at least one parameter in response to a difference between a sensor signal and a desired signal.

The method may further comprise controlling at least one of movement of the bucket of the ALV to at least one selected pose, and movement of the ALV relative to the rock pile. The method may further comprise perturbing an ALV velocity based on a sensor signal representative of interaction between the bucket and the rock pile during at least one previous dig, and modifying a sensor signal derived from boom and bucket force error measurement of at least one previous dig. The method may further comprise modifying a sensor signal being provided to a boom admittance controller, and modifying a sensor signal being provided to a bucket admittance controller, wherein modifying is based on feedback from at least one previous dig.

In one embodiment, controlling may include mapping a total force error to a range of possible sensed forces using at least one of proportional, integral, and derivative control. The method may further comprise mapping a signal from a previous dig to changes in dig controller response using at least one of proportional, integral, and derivative control.

In another embodiment, the pay load may be controlled based on a parameter of a breakout condition, or by modifying such a parameter.

Brief Description of the Drawings

For a greater understanding of the invention, and to show more clearly how it may be carried into effect, embodiments will be described, by way of example, with reference to the accompanying drawings, wherein:

Fig. 1 A is a schematic diagram of an ALV;

Figs. IB-ID are schematic diagrams of an ALV during three dig phases

corresponding to entry (Fig. IB), digging (Fig. 1C), and breakout (Fig. ID);

Figs. 2A and 2B are block diagrams of generalized dig controller embodiments;

Fig. 3 A is a block diagram of an admittance dig controller according to an embodiment;

Fig. 3B is a block diagram of an example of dig logic used in dig controller embodiments;

Fig. 3C is a block diagram of a dig controller according to one embodiment that includes an admittance controller;

Fig. 3D is a block diagram of a dig controller according to one embodiment that includes an iterative learning controller (ILC);

Fig. 3E is a block diagram of a dig controller according to another embodiment that includes an iterative learning controller;

Figs. 4A and 4B are block diagrams showing generation of boom and bucket correction forces according to embodiments;

Figs. 4C and 4D are block diagrams showing generation of entry throttle correction according to embodiments;

Fig. 5 is a plot showing dig efficiency points for 57 dig attempts using the experimental setup of Example 1 ; Figs. 6A and 6B are plots showing boom and bucket desired force profiles, respectively, and the target forces (in boxes) used by boom and bucket admittance controllers in an embodiment described in Example 2;

Fig. 7 is a plot showing desired and actual boom entry force rate of change, according to an embodiment described in Example 2; and

Fig. 8 is a plot showing boom and bucket desired forces used to calculate total error for each dig attempt (dark shading for negative error; light shading for positive error), for the embodiment described in Example 2. Detailed Description of Embodiments

As used herein, the term "autonomous loading vehicle" (ALV) is intended to refer generally to an autonomous, semi-autonomous, or robotic excavator machine or load-haul dump (LHD) vehicle used in accordance with the embodiments described herein.

As used herein, the term "actuator" is intended to refer to a component of the ALV that causes a change in vehicle configuration and/or motion. An actuator may carry out a function based on a command from a controller. For example, vehicle configuration may include position and/or orientation of a boom or dig tool, and/or position and/or orientation of the ALV.

As used herein, the term "actuated element" is intended to refer to a component of the ALV that is acted upon by an actuator, such as, for example, a boom or a dig tool, or actuator not currently receiving a command but being acted on by another actuator.

As used herein, the term "bucket" is intended to refer generally to a dig tool of an ALV, which may comprise a bucket, blade, chisel, fork, probe, bit, or other, as known in the art.

As used herein, the term "rock pile" is intended to refer generally to the material being loaded by the ALV. It is to be understood that the material may be of any type or composition as may be associated with excavation, construction, mining, and exploration, such as, but not limited to, soil, sand, gravel, ore, slag, salt, fragmented rock, regolith, or any combination thereof.

As used herein, the term "dig" is intended to refer generally to the actions performed by an ALV to carry out a desired function using its bucket. For example, a desired function may be to fill the bucket with material from the rock pile, wherein "dig" may be considered equivalent to "excavate". However, other actions (e.g., "load") may also be performed, and may optionally involve other dig tools. The dig actions of the ALV are controlled by dig controller embodiments described herein.

As used herein, the term "modify" means to change, adjust, or alter a magnitude or value, such as to increase or decrease a magnitude or value. The magnitude or value may pertain to a sensor signal. Modifying may be performed according to a mathematical operation or function, and/or may be performed in respect of a constant.

The dig controller embodiments for ALVs described herein provide efficient autonomous excavation in a wide range of materials in applications such as mining, construction, and exploration. The embodiments are particularly effective in rock piles including randomly sized fractured rock, which may be encountered in applications such as, for example, mining and construction.

A generic ALV is shown in Fig. 1A. Referring to Fig. 1A, the ALV includes a bucket 1 attached to a boom 2. The bucket is moved by actuating a bucket linear actuator 3 (curl), while the boom is moved by actuating a boom linear actuator 4 (hoist). These actuators, which may be electric, hydraulic, pneumatic, or a combination thereof, may be equipped with linear sensors or angular encoders to determine the configuration and/or motion of the bucket. Each actuator has a cylinder side and a rod side, shown as 7 and 8, respectively, for the bucket actuator 3. The boom and bucket actuators 4, 3, respectively, are connected to a vehicle 5 that can drive the boom and actuators to a desired location within the workspace. During loading, the vehicle drives the boom and bucket forward into the rock pile 6 (e.g., Fig. IB). The interaction between the bucket and the rock pile (e.g., Fig. 1C) causes changes in pressure on both the cylinder side 7 and rod side 8 of both linear actuators (e.g., Fig. ID), until the bucket is extracted from the rock pile.

Throughout the block diagrams of Figs. 2A, 2B, 3A-3E, and 4A-4D, descriptions of signals are provided in blocks with dashed lines. Generally, with reference to the block diagram of Fig. 2A, an ALV 10 interacts with a rock pile 6. Sensors produce sensor signals 14 representative of interaction between the bucket and the rock pile (e.g., reaction forces 40) and signals representative of motion of one or more bucket actuators 50. The sensors signals may be generated using one or more sensor or a combination of sensors selected from, but not limited to, accelerometer, force sensor, pressure sensor, torque sensor, load cell, and strain gauge. Bucket velocity may be sensed using one or more sensor or a combination of sensors, transducers, and the like selected from, but not limited to, accelerometer, linear variable differential transformer, wave reflection measurement (e.g., sonar, laser, infrared, video, optical encoder), and potentiometer (e.g., string, linear, or angular). The sensor signals are used by the dig controller 20, together with parameters 16 such as target forces 12, to generate control signals 18 that control the ALV.

Dig controller embodiments may include or utilize a sensing system 30 and controllers to control digging behavior of the ALV. Further detail is shown in the generalized embodiment block diagram of Fig. 2B. The sensing system 30 includes at least one sensor 32 and optionally a signal conditioner 34 that provides a sensor signal as input to the dig controller 20, which may include a logic device 22 and memory 24. Manual controls 26 and an operator interface 28 may also be provided. One or more sensors may be associated with an actuated element 64 of the ALV. For example, the controllers may include an actuator control device 60 to move the boom and bucket actuators 62 to an entry pose, to drive the ALV into the rock pile, and control forward motion of the ALV throughout the dig. The sensing system may detect that a force threshold is reached (e.g., 40 in Fig. 2A), upon which the dig controller 20 may use admittance controllers in an actuator control device 60 to regulate the velocity of the boom and/or bucket actuators 62 in response to the sensed forces. The sensing system may detect that the bucket actuator is fully extended (e.g., 50 in Fig. 2A), whereupon the forward motion of the ALV may be halted, and a position controller may be used to raise the boom to a weighing pose. The sensing system 30 may include at least one linear or angular sensor for each actuator (e.g., boom and bucket), and at least one force sensor for each actuator. In one embodiment, the force sensors include one or more pressure sensors on each actuator (e.g., one on the cylinder side, and one on the rod side of hydraulic actuators). The sensing system may optionally include a sensor for measuring the forward motion of the ALV. For example, the sensor may include one or more of an angular wheel encoder, an inertial sensor for detecting initial contact with the rock pile, and a vision system for detecting and/or assessing and/or characterizing the surface state of the rock pile. The vision system may include a ranging system capable of generating a 3-D representation of the rock pile surface. The 3-D representation may be used to select a point of contact between the bucket and rock pile such that digging time and effort are minimized. In these embodiments, a controller may include a proportional, integral, or derivative controller, or any combination thereof. Dig controller embodiments are shown in Figs. 3A-3E. The dig controller may include one or more admittance controllers 20A (Fig. 3A). Admittance controllers respond to changes in force with changes in velocity. Generally, an admittance controller seeks to maintain a mechanical admittance relationship between the environment (e.g., the rock pile) and a dig tool such that dig tool velocity is altered to achieve a desired environment reaction force. For example, in one embodiment, an admittance controller may map a force signal to a change in bucket motion (e.g., a desired velocity, as shown in Fig. 3C). Sensor signal input to the dig controller may be one or more parameter selected from, or may include all of: entry height, angle, boom force target, throttle, digging boom and bucket force targets, boom and bucket admittance controller gains, breakout condition, and weighing height and angle. For example, in one embodiment, when bucket forces increase, the velocity of the bucket is adjusted to bring the sensed forces within desired values. Use of admittance controllers provides embodiments that are relatively invariant to bucket-rock pile interactions because they regulate force, not position, of the bucket. This dynamic force regulation is particularly desirable for digging through a rock pile with random rock sizes, because pre-determined

(i.e., static) path targets would be difficult to follow given the randomly shaped obstacles that may be present in a typical rock pile. Admittance controller parameters may include proportional, integral, or derivative control terms, and a controller may implement a linear or nonlinear control scheme, e.g., according to a mathematical operation or function, and/or according to a constant. An admittance controller may be operated using dig logic 22 such as that shown in the embodiment of Fig. 3B.

Aggressiveness of an admittance controller may be governed by one or more parameters. In one embodiment, these parameters are the ALV entry throttle and the target force values 12 for the admittance controllers 20A for the boom and bucket. However, excavation efficiency is governed by the controller parameters and unknown rock pile parameters (more generally, the environmental parameters). The unknown rock pile parameters may include, for example, the rock size distribution, the pile shape, rock parameters (shape, Young's modulus, Poisson's ratio, etc.), moisture content, cohesion, and angle of repose, among others. It would be impractical to measure each of these parameters because of their number, and because the rock pile changes so frequently. While equations exist for modelling more homogeneous materials, such as soils, they are generally ill-suited for modelling non-homogeneous and highly variable targets such as fractured rock piles. The admittance controller overcomes this problem by treating the rock pile as an un- modelled body that provides changing reaction forces as the bucket passes through the pile. The admittance controller uses these forces to modify the motion of the bucket without explicitly knowing the characteristics of the rock pile. Whereas admittance controllers work well when the controller parameters have been tuned for a current state of the rock pile, they may need to be re-tuned when the rock pile changes significantly. For example, an admittance controller tuned for a wet rock pile may be too aggressive when the pile dries out, resulting in wasted effort and decreased efficiency.

In some applications or situations the boom actuator may be used to sense the digging force and provide a sensor signal that is used by the controller (e.g., an admittance controller, an adaptive admittance controller) to change the velocity of the bucket. Here, no commands are issued to the boom actuator, and forces sensed in the boom actuator are in response to the interaction between the dig tool and the rock pile. Thus, reaction forces received by the boom actuator provide an indication of the interaction between the dig tool and the rock pile. In this example the bucket actuator receives commands from the controller, and the boom actuator becomes an actuated element because no commands are sent to it by the controller. Alternatively, or in combination with the above, a sensor signal may be obtained by measuring strain in an actuated element, such as a boom.

It will be appreciated that one or more other elements of the ALV (i.e., other than the boom) could be used together with, or instead of the boom actuator, to provide sensor signal(s) to the controller, and used for controlling the ALV, provided that such one or more other elements are associated with appropriate sensor(s) to generate sensor signal(s) related to a dig parameter such as interaction between the dig tool and the rock pile.

In certain applications it may be desirable to optimize control of the ALV for less than maximum filling of the bucket. Such an embodiment may, for example, reduce strain on ALV components, thereby reducing down-time for maintenance and associated costs. This may be achieved by controlling the payload based on a parameter of a breakout condition, or by modifying such a parameter.

In a mining application, for example, rock pile parameters can vary significantly from one dig to the next even if the material being extracted remains of the same type. Admittance control as described herein has proved resilient to such changes; however, significant changes to digging conditions might give rise to a need to re-tune the admittance controller. Constantly tuning the admittance controller would not be practical or desirable. Some embodiments as described herein avoid the tuning problem by including at least one iterative learning controller (ILC) 70, 72, as shown in Figs. 3D and 3E. An ILC modifies the inputs to an admittance controller so that the controller parameters can remain constant while the controller response is altered. For example, as shown in Figs. 3D and 3E, an ILC may modify an input sensor signal, such as force, to an admittance controller, so that a desired dig behaviour is achieved while the entry throttle and force targets remain the same. In one embodiment, for example, Fig. 3E, the degree to which forces are modified is dictated by the force error history from previous dig attempts.

For example, if a dig attempt is more difficult than a previous attempt, the force error may be large and negative. In this case a large positive corrective force would be added to all forces going into the admittance controller, and the admittance controller would respond more aggressively. An advantage of an ILC is that it allows the algorithm to respond to changing rock pile conditions without having to re-tune (e.g., select constants that optimize performance) the admittance controller. This feature saves time, and eliminates the need for a specialist who would otherwise be needed for the re-tuning process.

In another embodiment, an adaptive admittance controller may be used. Parameters (e.g., proportional, integral, or derivative control terms) may be tuned or adapted dynamically (e.g., in real time or substantially in real time) to compensate for rapid changes in rock pile characteristics, such as stiffness, during a dig, thereby avoiding the need for modelling the rock pile. For example, an adaptive admittance controller may use the force tracking error to dynamically adjust admittance parameters throughout the dig in real time. In a further embodiment, an adaptive admittance controller may be used together with at least one ILC.

One embodiment of a dig controller, shown in the block diagram of Fig. 3D, includes two admittance controllers 20A and two ILCs 70, 72. A further embodiment may also include scripted entry and exit controllers. A further embodiment may include a detector for detecting if/when the ALV is stuck. The entry controller moves the boom and bucket actuators to an entry pose (e.g., bucket level with and just above the ground) using, for example, a proportional position controller. The ALV is then commanded to move towards the rock pile at a rate determined by the entry throttle set point, and the bucket engages the rock pile. When the sensed forces exceed a target value the admittance controllers begin moving the boom and bucket actuators. When the bucket has reached its maximum curl an exit controller takes over. The exit controller moves the boom and bucket to a weighing pose (e.g., bucket fully curled and raised above the rock pile) using, e.g., a proportional position controller. When the material in the bucket is weighed, the weight, dig time, and work performed by the actuators is used to assess the success of the dig attempt. An optimum dig maximizes bucket payload while minimizing dig time and work expended. In one configuration the admittance and position controllers operate at high frequency to perform the digging operations, while the ILCs only operate once per dig cycle.

An admittance controller may implement any mathematical relationship that maps the range of force errors to a range of possible actuator velocities. An admittance controller may modify a parameter, for example, in response to the magnitude of a sensed signal less a desired signal value. Fig. 3E is a block diagram of an admittance controller 20A according to one embodiment. Perturbed forces are used by the admittance controller to publish changes in the boom and bucket actuator velocities. These velocities are integrated to provide a set of desired positions for the boom and bucket actuator position controllers 60. The desired positions are tracked by the position controllers to provide the desired change in actuator length. The change in length causes the bucket to move in the rock pile, which causes the reaction forces to change. This change in force is sensed by pressure sensors 90 and used to calculate the new boom and bucket actuator forces 80. In one embodiment, these updated forces are again perturbed by the admittance ILC before being fed back to the admittance controller. Once a dig attempt is complete the total force error 85 is used to update the force perturbation for the next dig attempt, while the entry throttle ILC adds the new entry slope error to the previous entry slope errors so that the next entry throttle perturbation can be calculated.

In the embodiment of Fig. 3E, an ILC applies a correction 87 to the default entry throttle based on the entry force slope from several previous dig attempts. Other parameters used are the digging force targets. As shown in the embodiments of Figs. 4A, 4B, 4C, and 4D, the ILCs apply a correction 92 to the sensed forces 94 based on the total force error 96 from several previous dig attempts. For example, in Fig. 4C, while the entry throttle is initially tuned to a set value, the entry throttle is perturbed by the ILC to improve digging efficiency consistency. In Figs. 4C and 4D, the initial force rise for each dig attempt may be represented by the slope of a line passing through the lowest force reading, and the highest force reading, during the entry period (between bucket entry and admittance control). These slopes are compared against an ideal entry force slope to calculate the slope error for each dig attempt. These errors are stored in memory and a specified number n of them are summed. The sum 98 is used by the entry ILC to calculate how the entry throttle should be perturbed for the next dig attempt.

As shown in Fig. 4A, for example, an ILC may modify incoming forces so that the admittance controllers respond more aggressively. For example, Figs. 4C and 4D show that the ILCs increase the target entry throttle, and artificially increase the incoming forces. The increased values cause the ALV to enter the rock pile at a higher velocity, and curl and hoist the bucket faster. In these examples, increasing the entry, boom, and bucket velocities increases overall dig controller aggressiveness, and decreases digging variability compared to using parameters obtained from a training rock pile.

Dig controller embodiments may be implemented in analog and/or digital

(hardware/software) platforms. Specific implementations may be provided for compatibility with existing control systems, ALVs, sensors, etc., such as may be required to retrofit or upgrade existing systems and ALVs. For example, a dig controller may be implemented in whole or in part using discrete components, using digital technology (e.g., in a digital signal processor (DSP), field programmable gate array (FPGA), or application specific integrated circuit (ASIC) device), or using a combination thereof. One or more components of the dig controller may be implemented in an algorithm using a suitable hardware language such as, for example, very high speed integrated circuit (VHSIC), hardware descriptive language (VHDL), register transfer language (RTL), or Verilog. Such an algorithm may be implemented in, for example, a FPGA or ASIC device, or other suitable logic device. Some embodiments and implementations may include one or more sensors or transducers.

Embodiments will be further described by way of the following non-limiting Examples.

Example 1

This example illustrates the design and field testing of an embodiment of a loading algorithm based on admittance control using forces sensed from the bucket-rock interactions to modify the velocity of the bucket during digging. In this example, the loading algorithm (shown below) has three parts corresponding to three dig phases. The three dig phases, entry, digging, and breakout, are shown schematically in Figs. IB, 1C, and ID, respectively.

Loading algorithm:

1 : procedure ENTRY(pose, force)

2: while not in entry pose do

3: proportional controller moves boom and bucket

4: end while

5: while below entry force target do

6: drive forwards

7; end while

8: end procedure

9: procedure DlGGlNO( forces, extensions)

10; while not breakout condition and not stuck do

11 : continue driving into the pile

12; admittance controllers regulate actuator velocity

13; end while

14; while stuck do

15: shift to neutral

16: wait

17: shift to forward

18: end while

19: end procedure

20: procedure B R EA O UT(pose)

21 : while not in weighing pose do

22: proportional controller moves boom and bucket

23: end while

24; wait for steady state

25; weigh payload in bucket

26; end procedure

The entry phase is shown in Fig. I B. The entry phase ends when the bucket is in the entry position, and the forward motion of the ALV causes the bucket rock reaction forces to rise above a preset value. During the digging phase (Fig. 1C) the admittance controller causes the bucket to curl upwards or downwards to maintain a desired reaction force while the boom is used only to measure the digging reaction forces. The breakout phase (Fig. ID) starts when the bucket has fully curled, and ends when the bucket is in the weighing position The admittance controller is the part of the algorithm that governs the motion of the bucket through the rock pile. The admittance controller uses the error between the sensed dig reaction forces and a digging force target to alter the velocity of the bucket actuator. A generalized block diagram for the admittance controller is shown in Fig. 3C. Whereas any controller C can be used to map the force error to the actuator velocities, the admittance controller in this example is one-sided and proportional, such that

where VA is the actuator velocity, fa > 0 is the (admittance) proportional gain, and the force error is given by the target force fr minus the dig reaction force fs. For this loader and rock pile, fr was set at 80 kN, while fa was set at 1.1 x 10 ^"7 . These values were determined experimentally by adjusting them until the mass of payload in the bucket was high and consistent. The values may also be determined using off-line tuning methods for the admittance controller, wherein the values are calculated based on known vehicle parameters, and average rock pile stiffness. The controller was restricted to only making positive changes to the bucket velocity (i.e., only upward curl was allowed). In this example, this restriction was imposed so that no energy was wasted compressing rock against the underside of the bucket (and to maintain traction of the ALV wheels). However, this restriction is not always necessary, and may be omitted in other embodiments.

The drive train commands were set such that the loader was driven straight into the rock pile at a constant velocity. The entry position was selected such that the bucket scraped the asphalt substrate to ensure the bucket penetrated the rock pile at entry. The combination of drive train commands and entry pose resulted in substantially consistent penetration depth. After entry, the throttle was set to full to maximize the bucket actuator speed and power while the forward thrust was limited by applying partial brake. The brake level was set such that the forward thrust tended to increase the forces experienced in the actuators, which caused the admittance controllers to attempt to reduce the forces by curling backwards. The forward thrust tends to bias the controller towards the breakout condition ensuring that the dig completes before the bucket forces rise sufficiently to overcome the capacity of the actuators.

Field Test Apparatus and Method

An automated 1 -tonne surface loader and a blasted limestone rock pile were used to test and tune the loading algorithm. The loader was a Kubota R520s that was outfitted for automation by adding sensors, actuators, and on-board computer systems. Only boom and bucket extension and pressure sensors were used for this example. The boom and bucket actuator extensions were measured at 10 Hz by a custom hall effect sensor. Each extension sensor contained two Honeywell SPSL225 contactless IP69 linear encoders mounted in a custom housing, Two Measurement Specialties MSP-400 pressure sensors were installed on the rod and cylinder ports of each actuator so that the net force acting on the actuators could be calculated. The pressure sensor data was captured at 107 Hz by a single Arduino Uno board. The Arduino Uno pressure and actuator extension messages were passed to the main computer over a Robot Operating System (ROS) Electric network. The main computer was a Mini-ITX Intel Core i5 64-bit PC running Ubuntu 1 1.10, and ROS Electric. The main computer used a ROS network to publish and subscribe to topics over a wireless network. The autonomous loading algorithm was run on a separate Intel Core i5 64-bit laptop (running Ubuntu 1 1.10, and ROS Electric) connected to the wireless network. This laptop was also used for data collection.

tests, a fully-saturated controller was tested first, followed by an un-saturated admittance controller. One goal of these tests was to determine the difference between curling at maximum velocity and using the admittance controller to match curl velocity to the sensed reaction forces. Once the loading algorithm was tuned for this loader and rock pile, the various loading algorithm parameters were held constant for the dig attempts plotted in Fig. 5.

Each dig attempt started by manually moving the loader in front of the rock pile. The loading algorithm assumed control of the loader for entry, digging, and weighing. The loader was left in the weighing position for 5 seconds to allow the bucket and payload to stop moving. The payload mass mp was calculated using Equation (2), and the boom force fioom in the weighing pose. Equation (2) was determined experimentally by calibrating the loader using known masses. r p = A. oom - 3. IP ⁴

SL8 (2)

The volume of rock within the bucket was also verified by using the video feed from the wireless workstation. The payload was then dumped manually, and the loader was repositioned in front of the rock pile. Experimental Results

In total 57 dig attempts were made and the dig efficiencies from each dig are shown in Fig. 5. Of these, 23 dig attempts were conducted by curling the bucket at a maximum rate after entry (saturated digs), and 21 dig attempts were conducted by using a proportional (P) admittance controller, and 13 dig attempts were conducted by using a proportional-integral (PI) admittance controller, to match the bucket velocity to the sensed reaction forces

(controlled digs). Five saturated digs, two P controlled digs, and two PI controlled digs failed because the entry forces were insufficient for the saturated or un-saturated admittance controllers to take control of the dig. Saturated dig 18 (SI 8) and P controlled dig 1 1 (PI 1) took 30 seconds longer than average due to wireless network issues, and low rock pile entry respectively. The PI controller digs achieved 9% more payload and took 2 s longer than the fully saturated digs. Average saturated (S I 1), P controlled (P7), and PI controlled (PI68) dig attempts are presented, followed by the slow digs (SI 8 and PI 1), a high work dig (PI71), and two failed digs (S8 and P6).

When considering the actuator positions, valve commands (based on valve positions), and actuator forces for the nominal digs (SI 1 and P7), the valves on the loader have a deadband between ± 0.5. No fluid can flow to the actuators for any commanded valve positions within the deadband, hence any command within the deadband can be treated as zero valve displacement (a closed valve). The saturated dig curls the bucket at maximum velocity and the forces oscillate severely. In the controlled dig, the admittance controller alters the curl velocity in response to the changing forces resulting in less severe force changes.

For the slow digs (S I 8 and PI 1), the lack of actuator response to the full valve commands, and the high forces (well above both the entry and digging targets) in SI 8 indicated a network communication issue between the loader computer and the laptop running the loading algorithm. The force profile does show the level of force imparted to the pile by the drive train when the bucket stops moving, and the final payload mass indicates that the bucket was filled by the end of the dig. PI 1 is more interesting because the forces are very close to the 80 kN force target throughout the dig. When the forces rise above the target the admittance controller curls the bucket, which causes the forces to drop and allows the vehicle to penetrate deeper into the pile. When the loader stalls against the pile the forces rise, and another curl command is sent. The command and force histories for the failed digs (S8 and P6) show that the forces never rose above the entry force, so the bucket controller was never activated, and the digs failed. The forces were likely low due to the bucket hitting the ground and unloading the front wheels. Unloading the front wheels decreases the thrust and dig reaction forces.

Work and dig time were calculated between entry (after the entry force target is reached), and breakout (when the bucket curls past the breakout condition), "Instantaneous" work (only the current force reading is used) was calculated by using Equation (3), where Wd is the work performed by the actuators, -F/i and F _c are the hoist and curl forces in the boom and bucket actuators, respectively, and dh and d _c are the displacements for each actuator. Let n be the total number of sensor readings and let the subscript i denote the time index associated with each sensor reading. Thus, instantaneous work was calculated as:

W _d = - (3)

The average dig efficiency values are given for both the saturated and P controlled digs in Table 1. The dig time rose dramatically when the bucket was controlled by the P admittance controller compared to when the bucket was moved at its maximum rate.

However, all controlled digs were more consistent than the saturated digs, and the payload mass was increased by 10 %.

Table 1. Saturated and P controlled dig efficiency averages

Total work was calculated using Equation (4), where both the current force and last force readings are used in the total work equation: ∑ [( >M + · - «¾ .i i I + (Fci + F _c j+i) ^■ - d _e ,i+i \] i=l (4)

The results are presented in Table 2. Table 2. Saturated, P controlled, and PI controlled dig efficiency averages

All digs

Saturated Difference P Controller Difference PI Controller td [s 6.5+12.1 +277 % 24.5+11.3 -33 % 8.1 +5.0 w _d m] 24 458±8 495 +15 % 28 145+6 711 -3 % 27 287+14 511 [kg] 721+326 0.2 % 719±170 +1.0 % 792+335

Without failed diss

td [s 6.2+12.6 +315 % 25.7±l().l. -37 % 9.0+3.8

" a N in ^' 25 658+8 697 +15 % 29 551+1 929 +8 % 32 246+8 76G

[kg] ^' 784+254 -3 % 755+40 +22 % 924+85

Example 2

Key parameters for the entry and digging phases of an ALV may be identified as the entry throttle and the boom and bucket force targets. Desired force profiles, such as shown in Figs. 6A and 6B, may be used as basis for dividing a loading ILC into two parts: the entry ILC that governs the entry throttle; and the admittance controller ILC that modifies the sensed forces going into the admittance controllers.

Entry ILC

The entry ILC compared the slope of the boom entry force profile to the slope of the desired entry force profile as shown in Fig. 7. A force rise below 100 kN/s indicates that the rock pile provided less resistance than a training rock pile, while a force rise above this target indicates a more resistive pile. Compensation for a less resistive pile may be achieved by adjusting the entry throttle according to the relationship shown in Equation 5.

n

y¾ ^* €¾ entry slope

i=l

(5)

In Equation (5), Γ entry is any desired controller that maps the entry slope error e _en try slope to a throttle increment, referred to as the entry throttle correction CEntry throttle. The simplest controller is a proportional controller entry that linearly maps slope error to a throttle correction increment. is the number of dig cycles to consider, n is an optional weight applied to each entry slope error. This weight term can be used to bias the correction towards a desired set of entry slope error readings. For instance, the most recent entry slope errors are likely to best represent the current state of the rock pile. For example, a weight that exponentially decreases with respect to may be used to apply the largest weights to the most recent tests.

Equation 6 shows a specific instance of the entry ILC where slope errors from five dig attempts are multiplied by an exponentially decreasing weight, and summed before being multiplied by a proportional gain entry.

After five additional dig attempts the initial correction would be forgotten, and the current correction would be based on the five more recent entry slopes. A similar ILC is used to apply a correction to the sensed forces going into the admittance controllers.

Admittance ILC

The admittance ILC compares the calculated forces to the desired forces, and uses the result to modify the forces used by the admittance controllers. An example force profile is shown in Fig. 8. The total error between the sensed forces and the desired forces eFNet is calculated using Equation 7.

Again, the integral of the force difference (Foesired - Fsensed) is taken over the digging duration between tstart and few. The admittance ILC correction to the sensed forces C sensed farces is calculated using Equation 8.

Again, an exponentially decaying term is used to bias the correction towards the five most recent dig attempts, but any number of dig attempts can be considered, and all could have equal weight, y admittance is the admittance ILC gain. Any general controller Y admittance could be used instead of the proportional controller ) 'admittance.

An advantage of the ILCs is that once the admittance controller parameters are tuned for a given vehicle and rock pile they need never be tuned again, The ILCs discussed in this section have only two parameters each: the number of previous dig attempts /, and the ILC gains y. Another way to view the ILC gains is in terms of the aggressiveness of the entire digging algorithm (admittance and ILC controllers).

If the ILC gains are high the admittance controllers will respond more aggressively to changes in the rock pile, and if the ILC gains are low the controllers will respond less aggressively. This level of control is perfect for operators since it is a single value that can be tuned based on the overall loading goals. If an LHD payload is below the desired mass flow rate of the mill the operator can increase the aggressiveness of the controller by increasing the ILC gains. If the mass flow rate exceeds what the mill can handle, the ILC gains can be reduced, e.g., to save on tire wear and fuel consumption.

The contents of all references cited herein are hereby expressly incorporated by reference.

Equivalents

Those skilled in the art will recognize or be able to ascertain variants of the embodiments described herein. Such variants are within the scope of the invention and are covered by the appended claims.