Autonomous ground vehicle

Abstract

An autonomous ground vehicle (1) comprising a load-bearing structure (2) resting on the ground by means of ground contact (3, 4), at least one electric steering actuator (6) for steering the vehicle (1), at least one electric powertrain (5) for moving the vehicle (1), at least one sensing device (11) for an environment in front of the vehicle (1) with reference to a direction of travel of the vehicle, an electronic control system (7) for the steering actuator (6) and the powertrain (5), and a battery (12) for powering the vehicle (1), the electronic control system (7) comprising a navigation unit arranged to determine a path for the vehicle (1) to a destination point. The electronic control system includes a real-time controller implementing predictive control. FIGURE IN ABRIDGED DIAGRAM: Fig. 1

Description

Title of the invention: Autonomous land vehicle

[0001] The present invention relates to the field of autonomous vehicles.

[0002] BACKGROUND OF THE INVENTION

[0003] An autonomous land vehicle includes, for example, a support structure resting on the ground by wheels, and, in said support structure: an electric powertrain driving at least one of the wheels, at least one obstacle detection device in front of the vehicle, an electronic control unit and a vehicle power supply battery.

[0004] The electronic control unit provides navigation, vehicle guidance, and powertrain management. Powertrain management is generally achieved through closed-loop control using a proportional, integral, derivative (PID) controller or through adaptive control. In both cases, it is very complex to predict whether the battery has sufficient energy to allow the vehicle to overcome an obstacle in its path.

[0005] SUBJECT OF THE INVENTION

[0006] The invention is intended in particular to improve the control of autonomous land vehicles. Summary of the invention

[0007] For this purpose, the invention provides an autonomous land vehicle, comprising a load-bearing structure resting on the ground by means of ground connection, at least one electric steering actuator enabling the vehicle to be steered, at least one electric powertrain for moving the vehicle, at least one device for detecting an environment in front of the vehicle by reference to a direction of movement of the vehicle, an electronic control system for the steering actuator and the powertrain, and a battery for powering the vehicle.The electronic control system includes a navigation unit arranged to determine a vehicle route to a destination point and a real-time controller implementing predictive control from the following steps carried out at each current sampling instant of a series of sampling instants: - from a battery model and a vehicle model, determine, at the current sampling instant and at each future sampling instant of the series of sampling instants, a maximum electrical power supplied by the battery to the powertrain and deduce a maximum propulsive power produced by the powertrain taking into account the maximum electrical power; . determine, from a terrain model representative of at least one elevation of the ground in front of the vehicle with reference to the direction of movement of the vehicle, a path and an energy required to complete the determined path by summing the kinetic energy of the vehicle and the potential energy of the vehicle over the determined path, and deduce a power required over the determined path at the current sampling instant and at each future sampling instant of the series of sampling instants, the power required at each of these sampling instants being less than or equal to the maximum propulsive power along the path; develop the corresponding control of the powertrain and steering actuator; - if no route can be determined with the power required at each of these sampling moments less than or equal to the maximum propulsive power, stop the vehicle.

[0008] By journey, we mean the completion by the vehicle of a course at a predetermined speed, this speed not necessarily being constant throughout the entire journey. A predictive control method finds at each instant a finite set of future commands to apply to a system so that the future behavior of the system, over a finite prediction horizon, is optimal. The behavior is considered optimal with respect to the optimization problem posed to solve the predictive control when said behavior minimizes the cost function of the predictive control.Thus, with this invention, the predictive control method (or MPC, from the English "Model Predictive Control") takes into account the energy and power constraints imposed by the vehicle and the terrain to determine the vehicle's ability to complete the journey and, if not, to determine, if one exists, a new route compatible with the energy and power constraints to ensure that it can overcome or avoid any detected obstacles. The predictive control method will therefore provide the set of commands to be applied to the vehicle's actuators (powertrain and steering actuator) so that it completes the journey, takes a new, less difficult route, or abandons the journey if no less difficult route exists to reach the destination.

[0009] Other features and advantages of the invention will become apparent from the following description of a particular, non-limiting embodiment of the invention. Brief description of the drawings

[0010] Reference will be made to the attached drawings, among which:

[0011] [Fig-1] [Fig.1] is a schematic top view of a vehicle according to the invention;

[0012] [Fig.2] [Fig.2] is a block representation of the control system of the vehicle according to the invention; DETAILED DESCRIPTION OF THE INVENTION

[0013] With reference to [Fig. 1], the vehicle according to the invention, generally designated as 1, is an autonomous land vehicle comprising a load-bearing structure 2, such as a chassis and / or a body, resting on the ground by means of ground contact, namely, in this case, front steering wheels 3 and rear drive wheels 4, which are mounted on the load-bearing structure 2 to pivot about a substantially horizontal axis. The rear wheels 4 are mechanically connected to a powertrain 5 and the front wheels 3 are mechanically connected to an electromechanical steering actuator 6.

[0014] The powertrain 5 here comprises an electric motor and a mechanical and / or hydraulic transmission chain connecting an output shaft of the electric motor to the rear wheels 4 to transmit the movement from the output shaft to the wheels.

[0015] The electromechanical steering actuator 6 here comprises an electric motor having an output pinion meshing with a rack connected to the front wheels 3. Alternatively, the electromechanical steering actuator 6 comprises linear actuators or cylinders extending between the support structure 2 and the front wheels 3.

[0016] Vehicle 1 further comprises an electronic control system generally designated as 7 comprising - an electronic location unit comprising here a satellite signal receiver 8 (satellite positioning system of type GPS, GALILEO, GLONASS or BEIDU) and an inertial unit 9; - an electronic computing unit 10 comprising a processor, memory containing programs executable by the processor, and an interface for communicating with the vehicle actuators 1 and the electronic navigation unit for exchanging signals with them. These signals include commands sent by the electronic computing unit 10 to the actuators and status and / or operating data sent to the electronic computing unit 10 by the actuators and / or sensors associated with the actuators. These signals also include position information from the electronic positioning unit.

[0017] A detection device 11 is also connected to the interface of the electronic computing unit 10 and is mounted on the chassis 2 to detect the robot's environment. The detection device links a predetermined effective range defining a maximum coverage area around the vehicle 1, this area having a theoretical circular shape with a radius equal to the predetermined effective range of the detection device 11 (commonly referred to as the perception radius). The detection device 11 here comprises A laser remote sensing device (or "lidar"). Recall that such a detector comprises a laser emitter to emit successive incident laser pulses in all directions around vehicle 1 and a receiver for the laser pulses reflected by obstacles, allowing, through telemetry, the determination, for each orientation of the laser emitter, of the distance of vehicle 1 from the obstacles encountered by the incident laser pulses. The detection device 11 also includes a radar.Although the detection device 11 detects the environment of vehicle 1 in all directions, we are primarily interested here in the environment in front of the vehicle, relative to its direction of travel, here at a horizontal angle of 180° (the detection device could also be mounted to rotate and be oriented in the direction of travel of the robot; or include sensors oriented in opposite directions when the robot is configured to move in two opposite directions). The signals provided by the detection device 11 are representative: . - on the one hand, the bearing and distance of the obstacles detected by the detection device 11, and - on the other hand, the positive or negative elevation of the ground.

[0018] The electrical / electronic elements of the vehicle 1 (the actuators 5, 6; the electronic control system 7 and the detection device 11) are connected to a power supply circuit comprising a battery 12, here a lithium / ion battery, on board the vehicle 1.

[0019] The vehicle 1 further comprises a main braking device (not shown). This preferably includes a regenerative braking circuit allowing the electric motor of the powertrain 5 to be used as a generator to recharge the battery 12 and optionally an auxiliary braking device using eddy currents or mechanical friction. Each braking device is connected to the electronic control system 7 for control by the latter.

[0020] The programs executable by the processor of the electronic computing unit 10 include instructions implementing a control method according to the invention.

[0021] With further reference to [Fig. 2], according to the method of the invention, the electronic computing unit 10 comprises a real-time controller implementing a predictive control algorithm MPC and performs, according to a predetermined sampling period and over a predetermined prediction time horizon, an analysis of data from the elements connected to it. A predetermined path TD to a destination point indicated by a driver of the vehicle 1 was generated by the electronic control unit 10 via the electronic localization unit. Alternatively, the vehicle does not have a driver and the coordinates of the destination point are transmitted to the electronic control unit 10 via a radio transmitter / receiver, not shown here, connected to the electronic control unit 10.

[0022] At each current sampling instant of a series of sampling instants corresponding to the predetermined prediction time horizon, the electronic control system 7: - updates an MT terrain model from data from the detection device 11 including the slope of the ground in front of vehicle 1 and any obstacles possibly present in front of vehicle 1 (block A in [Fig.2]); - determines a mathematical representation of these obstacles to allow their inclusion in predictive control (block A); - from a battery model MB corresponding to battery 12 and a vehicle model MV corresponding to vehicle 1, determines, at the current sampling time, a maximum electrical power Pmax supplied by battery 12 to vehicle 1 (i.e. to the electrical / electronic elements of vehicle 1) and deduces a maximum propulsive power Pmot produced by the powertrain 5 taking into account the maximum electrical power; - determines, from the terrain model MT, a path to the destination point and the energy required to complete the determined path by summing the kinetic energy of vehicle 1 and the potential energy of vehicle 1 along the determined path, and deduces the required power P along the determined path at the current sampling instant and at each subsequent sampling instant in the series of sampling instants. The path is determined by the electronic control unit 7 such that the power required at each of these sampling instants is less than or equal to the maximum propulsive power Pmot, and generates the corresponding command for the powertrain 5 and the steering actuator 6 to complete the determined path.

[0023] If the electronic control unit 7 fails to determine a route that complies with said condition, the electronic control unit 7 stops the vehicle 1.

[0024] It is understood that the predictive control is based on a model of the terrain in particular in front of vehicle 1, on a model of battery 12 and on a model of vehicle 1.

[0025] The MB battery model is arranged to allow the maximum power that the battery 12 is capable of delivering and the remaining energy in the battery 12 to be known at any time. This model is, for example, that described in the document Chen Z., Fu Y., Mi C., “State of Charge Estimation of Lithium-Ion Batteries in Electric Drive "Vehicles using extended-Kalman Filtering," IEEE Transactions on Vehicle Technology, 62(3), 1020-1030, 2012. The maximum power Pmax that battery 12 can deliver is evaluated by taking into account the energy consumption of all the electrical / electronic components present in vehicle 1. The electrical / electronic components in vehicle 1 may have different energy consumptions depending on their use. It is assumed that the maximum available power is constant over the predetermined prediction time horizon (i.e., for the calculation of the current predictive control). To account for the efficiency of the motor and the transmission chain, the maximum motor power is determined using a vehicle model in the form: Pmot = a Pniax with 0 < a ≤ 1. The vehicle model MV is developed from the documentation of the vehicle's transmission chain components and / or experimentally.The coefficient α is determined by the successive efficiencies of the transmission chain. For example, if the energy flows along A -> B -> C, with the A -> B "link" having an efficiency f2i and the B -> C "link" having an efficiency ay, then the overall efficiency is α ~ aia2. The efficiencies αi and αi are generally provided in the manufacturers' technical documentation, but sometimes these data must be determined experimentally. Battery 12 provides the predictive control system with real-time data relating to available power and available energy.

[0026] The MT terrain model is here developed from the signals provided by the detection device 11. An example of a method for developing such a model is described in the document Larson J., Trivedi M., “Lidar based off-road negative Obstacle Detection and Analysis”, IEEE Conference on Intelligent Transportation Systems (ITSC), 192-197, 2011.

[0027] The terrain model MT includes the ground elevation in front of vehicle 1 and any obstacles in front of vehicle 1. The detected obstacles are represented so that they can be taken into account by the predictive control in order to develop a route that bypasses or avoids obstacles that vehicle 1 cannot pass through (such as trees, buildings, rocks, pedestrians, etc.). The representation, also called shaping, consists of finding at least one equation of the form f(x, y)<0 that represents a geometric shape separating the free space and the space occupied by the obstacle. For example, a tree can be defined by a circle with center yj and radius C. The predictive control will then have the constraint of not entering this circle: / \ 2 / \2 . . Alternatively, ellipses, etc., can be used. hyperplanes or polytopes instead of circles.

[0028] As previously stated, the terrain model is taken into account by the Predictive control to develop an optimal vehicle route 1. By "route," we mean a path and a speed along that path, which amounts to controlling both the powertrain 5 and the steering actuator 6. The route is said to be optimal in that it respects the energy constraint:

[0029] Efraj Edisp

[0030] This constraint ensures that the available energy Edisp for the electric actuator(s) / motor(s) is sufficient to complete the entire route predicted by the optimal path (Etraj). If this is not the case, an alternative path (for example, the same path at a lower speed, or a different path) that meets this condition is proposed if it exists (if no such alternative path exists, the vehicle stops). It is then guaranteed that vehicle 1 will not stop during the ascent (assuming that the prediction time horizon is long enough to encompass the entire ascent). The energy Edisp preferably takes into account the energy efficiencies of the various components in the transmission chain, in accordance with the vehicle model MV.

[0031] For simplicity, it is assumed here that the energy required by vehicle 1 along the determined path is limited to the sum of the potential energy and the kinetic energy. The kinetic energy is calculated directly using the classical formula Σc − 1^2, where m is the mass of the vehicle, assumed to be known, and v is the speed, a quantity known from the vehicle model. It can be calculated for the entire prediction time horizon. The potential energy Epp = mgz (m being the mass of the vehicle, S the acceleration due to gravity, and z the altitude) is calculated using the ground elevation data contained in the terrain model.

[0032] The power required at each instant can be approximated by p _ > ^pp with Tt. Te A corresponds to the variation between two sampling periods and T is the sampling period.

[0033] The following power constraint is then added to the predictive control problem:

[0034] P <pmof

[0035] To determine the control law to be used, the predictive control algorithm relies on an optimization algorithm (AO) that takes into account the crossing constraints (ground elevation) as a function of the available power and energy. The optimization is obtained using a classic cost function (FC) in predictive control, namely:

[0036] J^refprediction^ Q (ref.-prediction^ + ( order^ - order^1 R (order^ - order^

[0037] The weights Q and R are set empirically, ref. is the reference for The system state at time i, prediction, the prediction at time i, command, the command at time i, command1^ is the reference command at time i, and H is the prediction horizon, ref is a vector containing all the system state values, including the system's position, speed, orientation, wheel steering angle, etc. (alternatively, in more complex models, this can also take into account motor currents / voltages). prediction^ is a vector containing the same elements as the reference (to ensure homogeneous differences) and comes from the model used by the predictive control, here the bicycle model. We expect command j to be close to command^\ so that the system prediction is on ref. The commands here are the vehicle's speed and the wheel orientation.

[0038] Optimization allows us to determine the commands that will be applied to the system. The implementation of such a cost function is classic and will not be detailed further here (see for example the document Rawlings JB, Mayne DQ, “Model predictive control: Theory and design”, Nob Hill Pub. Madison, Wisconsin, 2009 and the document Pouilly-Cathelain M., “Synthesis of controllers adapting to multiple high-level criteria by predictive control and neural networks”, 2020).

[0039] Of course, the invention is not limited to the embodiment described but encompasses any variant falling within the scope of the invention as defined by the claims.

[0040] In particular, the means of contact with the ground may be wheels, tracks, skids, a hovercraft-type skirt, or other means, or a combination thereof. In the case of a wheeled vehicle, all wheels may be drive wheels and / or steering wheels. Steering may result from orienting the wheels or from differentially rotating them. The steering actuator may also act on a control surface to generate a directional aerodynamic force on the latter.

[0041] The vehicle may or may not be piloted, may or may not carry passengers.

[0042] The designation "powertrain" is to be taken in a broad sense and refers in Unlike any other type of land vehicle propulsion system, such as pure front-wheel drive, pure rear-wheel drive, all-wheel drive, or even hovercraft propulsion, the propulsion system can be hybrid, combining a combustion engine with an electric motor.

[0043] The detection device 11 of the vehicle 1 may include one or more of the following sensors: lidar, radar, sonar, visible camera, infrared camera, or any other suitable sensor for obstacle detection and / or ground tilt detection.

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] The MT terrain model can be obtained in different ways, for example by downloading it from a database using coordinates provided by the vehicle's navigation unit, or by processing or updating an existing three-dimensional map. The data provided by the detection device is used, for example, to update a map stored in memory accessible to the electronic control system. This memory may be part of the electronic control system itself or belong to a server that the electronic control system accesses via a network. The AO optimization algorithm used to solve the constrained optimization problem of predictive control can be, but not limited to: a constrained gradient descent algorithm, an active ensemble algorithm, an algorithm based on metaheuristics (DE, PSO, QPSO, ...), gradient-free algorithms (for example: the simplex algorithm; the Nerder-Mead algorithm; the constraints will be formulated as barriers for gradient-free algorithms). It is possible to use other models of the MV vehicle than the bicycle model and for example any other model presented in the document Polack P., D'Andréa-No vel B., De La Fortelle A., Menhour L., "Coherence between Modeling and Control Objectives for Autonomous Vehicles", 20th National Congress on Pattern Recognition and Artificial Intelligence (RFIA' 16), June 2016, Clermont Ferrand, France. It is possible to use a different battery model than the MB battery model mentioned above. It is possible to use models to take into account energy losses, particularly those due to friction or wheel slippage on the ground. It is possible to modify the cost function FC of the predictive control to limit energy consumption as follows: J = (ref - prediction; ) Q (ref - prediction; ) + (commanded?? - command; )TR (command'”? - command; ) + EnergyConsumed] The energy consumed: EnergyConsumed, is the energy required determined as described earlier. It is possible to vary Pmax according to a system operating mode (for example: economy / ecological mode or sport mode on the vehicle). If necessary, mechanical constraints for overcoming obstacles can be added, for example by using maximum values ​​for altitude gradient: ^z^nin and S^max are vehicle-dependent quantities (for example, the height of the undercarriage and / or the positioning of the vehicle's center of gravity).

[0056] The terrain model advantageously takes into account the nature of the soil, for example to estimate a risk of getting stuck or the vehicle's grip on the ground.

[0057] Constraints to ensure passenger comfort and guarantee the integrity of the vehicle and cargo (limits of vibration, lateral acceleration, and / or longitudinal acceleration for example) can also be introduced into the predictive control.

[0058] Constraints corresponding to the "physical" limits of the vehicle can also be taken into account by the predictive control: - speed maintained within an interval, - acceleration and / or deceleration maintained within an interval; - variation in acceleration and / or variation in deceleration maintained within an interval; - steering angle of the wheels maintained within a range; - steering angle variation maintained within a range.

[0059] We can dispense with the assumption that the maximum available power is constant over the predetermined prediction time horizon (i.e. for the calculation of the predictive control in progress) if we are able to predict the future consumption of all electrical / electronic components and then use this prediction in the predictive control.< / pmof>

Claims

1. Demands autonomous ground vehicle (1), comprising a load-bearing structure (2) resting on the ground by means of ground connection (3, 4), at least one electric steering actuator (6) for steering the vehicle (1), at least one electric powertrain (5) for moving the vehicle (1), at least one sensing device (11) for an environment in front of the vehicle (1) with reference to a direction of movement of the vehicle, an electronic control system (7) for the steering actuator (6) and the powertrain (5), and a battery (12) for powering the vehicle (1), the electronic control system (7) comprising a navigation unit arranged to determine a path for the vehicle (1) towards a destination point,characterized in that the electronic control system comprises a real-time controller implementing predictive control from the following steps performed at each current sampling instant of a series of sampling instants: - from a battery model and a vehicle model, determine, at the current sampling instant and at each future sampling instant of the series of sampling instants, a maximum electrical power (Pmax) supplied by the battery (12) to the powertrain (5) and deduce a maximum propulsive power (Pmot) produced by the powertrain (5) taking into account the maximum electrical power; - determine, from a terrain model representative of at least one elevation of the ground in front of the vehicle (1) with reference to the direction of movement of the vehicle, a path and an energy required to complete the determined path by summing a variation in kinetic energy of the vehicle (1) and a variation in potential energy of the vehicle (1) between two sampling instants on the determined path, and deduce a power required (P) on the determined path at the current sampling instant and at each future sampling instant of the series of sampling instants, the power required at each of these sampling instants being less than or equal to the maximum propulsive power (Pmot) along the path; - develop the corresponding control of the power-propulsion unit (5) and the steering actuator (6); - if no path can be determined with the power required at each of these sampling instants less than or equal to the maximum propulsive power (Pmot), stop the vehicle (1).

2. Vehicle according to claim 1, wherein the detection device (11) is arranged to reveal the presence of an obstacle and determine a ground elevation in front of the vehicle (1) with reference to the direction of travel of the vehicle; and the terrain model is developed from data provided by the detection device (11).

3. Vehicle according to claim 2, wherein the data provided by the detection device (11) is used to update a map stored in a memory to which the electronic control system (7) has access.

4. Vehicle according to any one of the preceding claims, wherein the vehicle model is the bicycle model.

5. Vehicle according to any one of the preceding claims, wherein the predictive control relies on a constrained optimization algorithm such as a constrained gradient descent algorithm, an active ensemble algorithm, a metaheuristic-based algorithm, or a gradient-free algorithm.

6. Vehicle according to claim 5, wherein the constrained optimization algorithm implements a cost function J = L" ( ref, - predictionf Q ( refprediction^ + (coiwtwndep - command^)1 R (command^ - command^) with R and Q empirically set weights, ref a reference to a vector containing a set of system state values ​​at sampling time i, prediction, the prediction at sampling time .H the prediction horizon, and command{ the command at sampling time i.

7. Vehicle according to claim 6, in which the cost function is J = V. J ref. - prediction,] Q ( ref^ - prediction^ + ( command^ - command^ )TR ( command^? - command^ + EnergyConsumed with EnergyConsumed the energy required to complete the journey.

Images