WO2017120109A1

WO2017120109A1 - Vector in guidance out processing engine for autonomous vehicles

Info

Publication number: WO2017120109A1
Application number: PCT/US2016/069417
Authority: WO
Inventors: David Wayne RUSSELL
Original assignee: Individual
Current assignee: Individual
Priority date: 2016-01-06
Filing date: 2016-12-30
Publication date: 2017-07-13
Anticipated expiration: 2018-07-06

Abstract

In an autonomous vehicle, the architecture of the flight control/piloting system can be greatly simplified if the processes are broken down hierarchically. This disclosure teaches the utilization of learning systems such as neural networks, multiprocessor deep learning arrays, or genetic programming algorithms to implement a hierarchical mixture of experts processor which receives a navigation data structure and platform sensor information to create absolute or relative control signals for one or more motor control output channels at high speeds.

Description

SPECIFICATION TITLE OF INVENTION

Vector In Guidance Out Processing Engine for Autonomous Vehicles INVENTORS

David Wayne Russell, (USA) Winter Garden, Florida USA CROSS-REFERENCE TO RELATED APPLICATIONS US 62/275,673 1/6/2016

STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

Not Applicable

REFERENCE TO SEQUENCE LISTING, A TABLE, OR A COMPUTER PROGRAM

LISTING COMPACT DISK APPENDIX

Not Applicable

FIELD

[0001] The field of this invention relates generally to computational engines and more specifically to flight control processors for autonomous vehicles.

BACKGROUND

[0002] In most attempts at designing an autonomous vehicle, the approach is to design in software an analog of the processing performed in a human brain in performing the same actions, under the premise that if humans do it well the machine should work in the same way. Human brains, however, and electronic processing systems have fundamental differences which makes this premise difficult to realize at best.

[0003] The software version of the control system becomes significantly more complex when it tries to solve all of the problems associated with vehicle autonomy at once. In addition, general purpose computers are essentially designed to do everything poorly. They are capable of doing anything, but the inevitable consequence of that flexibility is that it does not do any one thing particularly well. Hardware accelerators have long been shown to be successful in implementing specific problem solutions at very high speeds.

[0004] In many piloted "drone" systems, the flight controller combines sensors and motor control to maintain stability of the platform independent of the pilot's directional control inputs. Motor and control surface controls, however, are often non-linear problems by their very nature, and software systems are either slow at non-linear problem solving or only attempt to

approximate non-linear solutions with linear solution algorithms.

[0005] A system is needed where flight control can significantly simpler and faster than an algorithmic aggregation of software programming.

BRIEF SUMMARY OF THE INVENTION

[0006] The Vector-In Guidance-Out (VIGO) engine is essentially a multi-stage parallel processing system implementing a non-linear control system. These controls handle the nonlinear nature of the system by implementing hardware versions of non-linear control techniques such as but not limited to fuzzy logic, finite state machines, neural networks, and/or genetic learning algorithms. Neural networks, fuzzy logic, and finite state machines have been studied and found applicable to the non-linear control systems common in UAVs.

[0007] What is unique is the combined application of these algorithms in a hardware

architecture. Just as the overall UAV flight problem is more easily solved when it is broken into a layered strategy where each layer is optimized for the computing environment and platform it represents, so too the low level flight dynamics are simplified by a hierarchical mixture of experts architecture.

BRIEF DESCRIPTION OF DRAWINGS

[0008] The detailed description is described with reference to the accompanying figures. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The use of the same reference numbers in different figures indicates similar or identical components or features. [0009] The novel features believed characteristic of the illustrative embodiments are set forth in the appended claims. The illustrative embodiments, however, as well as a preferred mode of use, further objectives, and features thereof will best be understood by reference to the following detailed description of illustrative embodiments of the present disclosure when read in conjunction with the accompanying drawings, wherein:

[0010] FIG. 1 shows a block diagram of data flow for a Vector- In Guidance-Out control processor.

DETAILED DESCRIPTION OF INVENTION

[0011] The following detailed description illustrates embodiments of the invention by way of example and not by way of limitation. The description clearly enables one skilled in the art to make and use the disclosure, describes several embodiments, adaptations, variations,

alternatives, and use of the disclosure, including what is currently believed to be the best mode of carrying out the disclosure. The disclosure is described as applied to an exemplary embodiment namely, systems and methods of autonomous a vector-in guidance-out engine for autonomous systems. However, it is contemplated that this disclosure has general application to vehicle management systems in industrial, commercial, military, and residential applications.

[0012] As used herein, an element or step recited in the singular and preceded with the word "a" or "an" should be understood as not excluding plural elements or steps, unless such exclusion is explicitly recited. Furthermore, references to "one embodiment" of the present invention are not intended to be interpreted as excluding the existence of additional embodiments that also incorporate the recited features.

[0013] One of the most non-linear flight dynamics problems is the response of the platform to stepwise changes in engine thrust and/or direction. In one embodiment only engine thrust is varied, in others engine thrust and direction are variables, or a combination of the two depending on phase of flight and situational variables.

[0014] In one embodiment fuzzy logic is utilized, with a separate processor for each engine. Depending on the platform, one to N engines might be implemented, although in common UAVs N tends to be eight or less. This is the lowest level of the VIGO implementation, outputting direct control electrical signals to the motor power control. In another embodiment the fuzzy logic controllers could be paired or grouped into other subsets of the total N. In other embodiments the fuzzy logic controllers might be implemented in other techniques as above.

[0015] The higher level of the implementation is then implemented in a learning system such as neural networks or genetic algorithms. This allows for a highly complex combination of both learning and non-linear algorithms, gaining the strengths of each while avoiding the weakness of each.

[0016] For example, the higher level system is responsible for receiving a magnitude and direction vector from the platform navigation system and comparing that to the platform's current attitude, speed, motor settings, and any other factors that can be sensed or inferred from the platform itself. It then outputs a limited number of analog signals or digital numeric values to control the N motors to achieve the platform's transition from state T to state T+l .

[0017] Learning algorithms are not well used in environments with output values that must vary over time or have a large number of outputs, but they do well when a few outputs are required against a large number of inputs. The learning system works well in finding an efficient transfer function from the information in the navigation vector to simple vectors for the motors.

[0018] As this transfer function is different for every type of platform, there is little to be gained by programmatic approaches and a dedicated learning system based on known profiles is sufficient and robust. In other embodiments the learning system may be modified based on the given mission profile and performance specifications of the platform.

[0019] Similarly, the fuzzy algorithms and/or state machines that make up the lower level do well with making quick output decisions but need few outputs to be easily implemented. By combining both techniques in a hierarchy both subsystems perform to their maximum ability and speed. In other embodiments a standard computing platform could be use for either system with similar programming, one familiar with the art would realize that this would simply slow the system's response and in some cases this might be acceptable, but it would not change the underlying invention of utilizing the hierarchical mixture of experts in realizing the VIGO functionality.

[0020] Processing is also simplified by the sheer speed of the system execution. While a UAV with standard programming systems might update its positional and control settings 100 times per second, a hardware navigation system might issue a new vector at a rate of around 1,000,000 control vectors per second. This means that the changes that VIGO has to deal with are very small, and the focus becomes how to make very fast small changes rather than how to control the motors and control surfaces for large changes over longer periods of time.

[0021] This focus on fast, small changes significantly reduces overshoot, undershoot, oscillation, and other challenges to standard control systems. In one embodiment the system update frequency is tuned to a rate that keeps the motor control units in or near a linear range of operation, simplifying the underlying control layer even further.

[0022] In one implementation, VIGO shares overall platform data such but not limited to altitude with the navigation system and takes advantage of the high data rate provided by navigation's sensor interpolation and predictive processing. In other implementations VIGO only utilizes the platform specific information from its own sensor suite, although these may also use predictive sensor interpolation to achieve high data rates.

[0023] The output to each of the motors is a simple vector - magnitude (RPM) and optionally X and Y offset of the motor or directional vanes attached to impact the motor air stream. This allows a very simple mathematical transform from the current to the next vector to be implemented with minimal hardware.

[0024] In one embodiment VIGO, employs predictive algorithms to anticipate the next vector when necessary. The current vector Tn is compared against some number of previous vectors to establish rates of change through successive derivatives. This is possible because the error, if there is a sudden change in the vectors from the fast navigation processor, is not large and can be compensated for in the next sample because of the high speed of the processors.

[0025] In another embodiment, because of vectored thrust, the platform does not have to be oriented with respect to the direction of flight. In this case the vector passed from navigation to VIGO also contains three angle rotational factors for the platform body with respect to the direction of flight which are also implemented by VIGO's vectored thrust motors, control surfaces, or both. [0026] Depending on the platform implementation, VIGO may need to develop control values for controls such as but not limited to motor RPM, vectored thrust gimbal offsets, vectored thrust rotation, control surfaces, vanes, safety devices, and terminal guidance.

[0027] In Figure 1, the input to the system is a vector 100 which is made up of speed

(magnitude) and direction X, Y, Z components. In addition the vector may contain attitude offsets such as but not limited to X, Y, and Z with respect to the XYZ of the input vector or in other embodiments with respect to the inertial measurement coordinate system (gravity).

[0028] The Vector Translation Processor (VCT) 105 receives the vector, inputs from the platform sensors 110 and stored platform variables 115. In one embodiment the VCT is implemented in hardware such as but not limited to a gate array, field programmable gate array, custom logic IC, or Complex Programmable Logic Device. In another embodiment it might be implemented in a standard programmable CPU and software, if the system does not need to be high speed.

[0029] In another embodiment it might be a combination of the two, with the software system monitoring slow varying signals and hardware calculating the high data rate outputs, or the software system used to monitor the hardware.

[0030] One additional input to the VTP is the Oscillation Overthruster (OOT) 120. Even at the high speeds of the Nav system and VIGO, it is possible for oscillations to develop. The OOT monitors the platform sensors to detect whether oscillations are occurring. This may be implemented in a hardware Fast Fourier Transform or CPU software system if the oscillations are low enough in frequency or the damping can be applied relatively slowly. The OOT introduces an oscillation into the VTP 180 degrees out of phase with the detected oscillation to dampen it much the way that noise-cancelling headphone operate.

[0031] The outputs of the VTP are N outputs, analog or digital, that specify the relative change to make in each of the N motors 135. The motors may be simple RPM based controls, voltage, current, PWM or they may also implement vectored thrust by changing the X,Y orientation of the motor and/or associated thrust vanes, ducts, or other surfaces. In one embodiment a fuzzy logic motor translation process directly converts the VTP output to motor values and X,Y relative changes. The motor position is the preferred embodiment are driven by voice coil actuators which are much faster and more immune to temperature changes.

[0032] The positioning ability can be achieved with two gimbal mounts or in another embodiment the positioning ability might be implemented with one gimbal and rotating the motor housing. This will have an effect on the motor translation process. Once the motor translation process has output a new motor RPM value, the motor driver circuit 130 alters the current, voltage, or other settings and control to the motor to achieve the desired result.

[0033] In another embodiment changes might be in an absolute coordinate and variable system rather than relative. The X,Y voice coil actuators 140 may also require a servo or other feedback system to lock into the correct position as given.

[0034] While the foregoing written description of the invention enables one of ordinary skill to make and use what is considered presently to be the best mode thereof, those of ordinary skill will understand and appreciate the existence of variations, combinations, and equivalents of the specific embodiment, method, and examples herein. The invention should therefore not be limited by the above described embodiment, method, and examples, but by all embodiments and methods within the scope and spirit of the invention. Further, different illustrative embodiments may provide different benefits as compared to other illustrative embodiments. The embodiment or embodiments selected are chosen and described in order to best explain the principles of the embodiments, the practical application, and to enable others of ordinary skill in the art to understand the disclosure for various embodiments with various modifications as are suited to the particular use contemplated.

[0035] The flowcharts and block diagrams described herein illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various illustrative embodiments. In this regard, each block in the flowcharts or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function or functions. It should also be noted that, in some alternative implementations, the functions noted in a block may occur out of the order noted in the figures. For example, the functions of two blocks shown in succession may be executed substantially concurrently, or the functions of the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.

Claims

The invention claimed is:

1) A system which receives a control data structure and converts it to one or more control signals via a hierarchical mixture of experts processor.

2) The system of 1 where the control data structure relates to platform navigation and/or trajectory data.

3) The system of 1 where the control signals drive some combination of platform

implemented physical actuators such as but not limited to motors, control surfaces, vanes, ducts, linear or rotational actuators, stepper motors, or voice coil actuators.

4) The system of 1 where the navigation data structure would also include platform

rotational offsets including some combination of the X, Y, and Z axes relative to the direction of flight.

5) The system of 1 where motor drive control and motor rotation and/or translation controls are included to produce vectored thrust from the motors.

6) The system of 1 where a learning system such as but not limited to neural networks, multiprocessor deep learning arrays, or genetic programming algorithms are used to implement a hierarchical mixture of experts processor which receives the navigation data structure and platform sensor information to create absolute or relative control signals for N motor control output channels.

7) The system of 1 where a computer CPU and software are used to implement a

hierarchical mixture of experts processor, which receives the vector and platform sensor information to create absolute or relative control signals for N motor control output channels.

8) The systems of 1 where the vector translation processor also receives some combination of stored platform variables to calculate the output control channels.

9) The system of 1 where the vector translation processor also receives input from an

oscillation overthruster component which detects oscillations in the platform and produces a dampening signal.

10) The system of 1 which implements a non -linear control system in hardware such as but not limited to fuzzy logic and/or finite state machines to convert a motor speed and optionally direction information to motor drive values. 11) The system of 1 which implements a non -linear control system utilizing a CPU and software to convert a motor speed and optionally direction information to motor drive values.

12) The system of 1 which implements a non -linear control system in hardware using a combination of CPU, programming, and customized hardware to convert a motor speed and optionally direction information to motor drive values.

13) The system of 1 where voice coil actuators are used to provide high speed control of the motor vectored thrust.

14) The system of 1 where the oscillation overthruster is comprised of some combination of CPU and programming, hardware Fast Fourier Transform, and/or customized hardware.

15) The system of 1 where the platform sensors are comprised of some combination of control parameters such as but not limited to attitude, speed, fuel remaining, control surfaces, servo feedback, and other internal and external sensors as determined by the mission profile.

16) The system of 1 applied to air, surface, sea, or subsea vehicle performance.

17) The system of 1 applied to general non-linear control systems.