Drive control method for a mobile robot

Abstract

The aim is a drive control method for a mobile robot (1) with a multiplicity of wheeled legs (3) with non-steerable wheels (8), which is more efficient and reliable. The problem is solved with the steps: a) by means of a planning software (111) on a computer hardware (11): - accessing input provided via a user interface and comprising a goal, and - determining a plan depending on the goal and environmental data maps, b) by means of an execution software (101) on the electronics (10) of the mobile robot (1): according to the plan, operating the hip joint (4), the knee joint (6), and the wheel motor (9) of every wheeled leg (3); and c) real-time controlling by means of the execution software (101) and the planning software (111).

Description

[0001] Drive control method for a mobile robot

[0002] TECHNICAL FIELD

[0003] The invention relates to a drive control method for a mobile robot according to preamble of claim 1, a use of a neural network for a drive control method for a mobile robot and a mobile robot.

[0004] STATE OF THE ART

[0005] US2021276642 discloses a modular robotic vehicle or (MRV) including a chassis and body having any shape and dimension to include an enclosed cab in which passengers are seated therein or a passenger to ride on a seat without an enclosed cab. The vehicle's modular chassis further comprising leg array rotatably connected therein, the leg array including actuators causing flexing and bobbing motion for keeping the MRV stabilized when traversing over various ground surfaces in indoor or outdoor environments. The leg array providing walking and steering capability allowing the MRV to transverse during a navigation mode, the wheel providing differential steering propulsion or braking capability, such that the wheel operates like a foot when powered off during a walking mode and rotates when powered on during a drive mode, the MRV to transport passengers and / or cargo.

[0006] US2022097785, CN112874651, CN111846002, CN110861728 and the article XP011789464: Trajectory Optimization for Wheeled-Legged Quadrupedal Robots Driving in Challenging Terrain in IEEE Robotics and Automation Letters by Medeiros, Vivian S. et al. are advancements of the last years in the field of mobile robots, which rely on different drive technologies and methods. The closest prior art is XP011789464. It however does not describe a sufficiently reliable and efficient method to drive a mobile robot with non-steerable wheels. The prerequisites for the steerability of the wheel, e.g., via an ankle joint, are often extensive and complex hardware solutions. The same holds for many fauna-inspired realizations of the rotatable connection between the leg and the body, such as for example in the form of a ball joint. Also, standard solutions for the actuation of the leg yield a limited set of directly provocable motions. Moreover, concerning the operation of the robotic vehicle, fast, smooth, and efficient movement can hardly be achieved by manual control and automatic, standard, preprogrammed responses of the robotic vehicle to pre-defined events.

[0007] Most known robotic vehicles also deploy either omnidirectional wheels or a steering motor for each wheel. The former adds complexity to the wheel design, and omnidirectional wheels are known to only properly function on flat surfaces with good ground interactions. The latter adds additional hardware parts because one more motor per leg is needed to steer the vehicle. The additional steering motor increases the inertia of the leg and negatively affects performance and costs.

[0008] Often, the wheels are also added to the base. In this case, the robot can only use the wheels on fully flat surfaces, and the wheels' mass is added directly to the main body, which decreases usable payload.

[0009] So far, mobile robots with wheeled legs, where the wheels are nonsteerable, cannot be controlled in an efficient and reliable way.

[0010] DESCRIPTION OF THE INVENTION

[0011] It is an objective of the invention to provide a mobile robot with a multiplicity of non-steerable wheels, a drive control method for a mobile robot and a use of a neural network for a drive control method for a mobile robot in general, such that the mobile robot can move omnidirectionally, fast, smoothly, and efficiently and overcome challenging terrain with minimal hardware. Variations of feature combinations and minor adaptations of the invention respectively can be found in the detailed description, are illustrated in the figures and have been included in the dependent claims.

[0012] BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Further understanding of various aspects of the invention can be obtained by reference to the following detailed description in conjunction with the associated drawings, which are described briefly below.

[0014] It should be noted that in the differently described embodiments, the same parts are provided with the same reference symbols or the same component names, the disclosures contained in the entire description being able to be applied analogously to the same parts with the same reference symbols or the same component symbols.

[0015] A preferred exemplary embodiment of the subject matter of the invention is described below in conjunction with the attached drawings.

[0016] Figure 1 shows a perspective view on a mobile robot with a payload interface.

[0017] Figure 2 shows a perspective view on a mobile robot with one wheeled leg in a stepping position.

[0018] Figure 3 shows a top view on a mobile robot with one wheeled leg in a stepping position.

[0019] DESCRIPTION

[0020] Automated vehicles, like mobile robots 1, become more and more important, especially in the field of transportation of goods and people, but also e.g., for the exploration of unknown terrain. For the use of such an automated vehicle to be advantageous compared to traditional solutions, the former needs to be fast, efficient, and it needs to be able to overcome obstacles like stairs, tilted, wet or icy surfaces, uneven undergrounds like lawns, pebble stones or even boulders, and traffic. This on the one hand bears the challenge of providing hardware which is as simple as possible regarding operation and production, is as little error-prone as possible, has itself as little weight as possible, and does allow for a physical model for the movement of the vehicle which is as simple as possible. On the other hand, highly capable software, computer devices and electronics for a drive control method to operate the automated vehicle, which can pursue given goals and at the same time react in real-time to events influencing the state of the vehicle, needs to be provided. Moreover, the drive control needs to optimally exploit the properties of the hardware and vice versa. This was achieved here as described below.

[0021] The mobile robot 1 presented here is exemplarily shown in Fig. 1 to 3.

[0022] Fig. 1 shows a mobile robot 1 with a torso 2 and four wheeled legs 3. Each wheeled leg 3 comprises a hip joint 4, an upper leg 5, a knee joint 6, a lower leg 7, and a wheel 8 with an attached wheel motor 9. For clarity, in Fig. 1 the reference signs corresponding to the wheeled legs 3 are shown only for two of the wheeled legs 3. The mobile robot 1 further is equipped with at least one energy source, at least one sensor, a communication system, and electronics 10. The electronics 10 are operatively connected to the at least one energy source, the at least one sensor, the communication system, and to the hip joint 4, the knee joint 6, and the wheel motor 9 of every wheeled leg 3. Further equipment and additional wheeled legs 3 are possible.

[0023] The energy source is here a battery, which is operatively connected to the electronics 10. Alternatively, the energy source can for example be an internal combustion engine combined with a fuel tank and an energy transformer which has as output electrical energy and is connected to the electronics 10, or a combination of such an engine and a battery.

[0024] The torso 2 is here shaped as a shallow elongated cuboid with recesses and protrusions. At each of the two long, narrow side faces of the shallow cuboid, one of the wheeled legs 3 is attached within each of the two outer longitudinal quarters of the side face. More precisely, the torso 2 is connected to the hip joint 4 of each of the four wheeled legs 3, which in turn is connected to the upper leg 5 of the corresponding wheeled leg 3. Other shapes of the torso 2 and other arrangements of the wheeled legs 3 are possible.

[0025] The hip joint 4 of every wheeled leg 3 is realized here by means of a hip joint component 41 of the form of a uniform prism. It comprises two revolute joints with mutually perpendicular axes, where the axes fall in a line with a first and a second connection protruding from the hip joint component 41. Alternatively, the axes can be arranged in a different angle with respect to each other, as long as they are not parallel. The first connection protrudes centrally from and perpendicular to one of the bases of the prism of the hip joint component 41. Its distal end, with respect to the hip joint component 41, is attached in a stationary way to the inner side face of the corresponding upper leg 5 within the uppermost longitudinal quarter of the latter, and its proximal end is attached to the hip joint component 41 in such a way that the first connection can rotate with respect to its central longitudinal axis, i.e., with respect to the direction of its extension. The revolute joint corresponding to the first connection thus allows the corresponding wheeled leg 3 to rotate in a plane p_l roughly parallel to the bases of the prism of the hip joint component 41. Note that roughly parallel and roughly perpendicular is understood here in the sense that the plane may have an angle of at most fifteen degrees with respect to a truly parallel and truly perpendicular plane respectively at the axis of rotation and in any direction. Furthermore, note that more precisely, different points of the wheeled leg 3 may move in different, roughly parallel planes p_l.

[0026] The second connection protrudes from the hip joint component 41 along a central axis parallel to the bases of the prism of the latter. Its proximal end, with respect to the hip joint component 41, is attached to the hip joint component 41 in a stationary way, and its distal end is attached to the torso 2 in such a way that the second connection can rotate with respect to its central longitudinal axis, i.e., with respect to the direction of its extension. The revolute joint corresponding to the second connection thus allows the wheeled leg 3 to rotate in a plane p_2 roughly perpendicular to p_l. Naturally, the scope of this rotation is limited by the torso 2 which crosses this plane. Note that more precisely, different points of the wheeled leg 3 may move in different, roughly parallel planes p_2.

[0027] Other realizations of the hip joints 4 are possible, e.g., with a hip joint component 41 of alternative shape, without a hip joint component 41, with different arrangements of the revolute joints concerning their relative distance and angle, or with different arrangements for the attachment to the torso 2.

[0028] The knee joint 6 of every wheeled leg 3 is realized here as a revolute joint allowing for changes of the angle between the corresponding upper leg 5 and the corresponding lower leg 7 roughly within the plane p_l. Alternative possible realizations of the knee joints 6 include sliding joints, which also allow for a shortening of the distance from the top of the upper legs 5 to the corresponding wheels 8. The hip joint 4, the knee joint 6 and the wheel motor 9 of every wheeled leg 3 are preferably electric. Alternatively, one or more of them can be hydraulic or pneumatic.

[0029] The wheel 8 of each wheeled leg 3 is here attached to the corresponding lower leg 7 in such a way that its axis of rotation is fixed to lie in a plane roughly parallel to the support of the wheel 8. The wheel motor 9 of the wheel 8 is preferably attached to the wheel hub of the latter and is used to accelerate, i.e., adjust the rotation speed of the wheel 8. In particular, there are no motors to steer each wheel 8, i.e., there is no steering mechanism at the wheels 8, and therewith the wheels 8 are nonsteerable. The position of the wheels 8 in the cartesian x, y and z direction is allowed to be controlled by the joint architectures and the three-dimensional active suspension of the wheeled legs 3. In other words, the axis of rotation of the wheel 8 is fixed with respect to the lower leg 7. Swivelling movements of the non-steerable wheels 8 are not provided for or can be changed by a mechanism. In this application, we assume mobile robots 1 with such non-steerable wheels 8 for which optimised control is to be achieved. This means that further mechanical controls and adapted wheels to be controlled do not play a role. The achievement here is the software-guided solution.

[0030] The torso 2 contains here the electronics 10 and the battery, where the latter is operatively connected to the electronics 10. The electronics 10 comprise a microchip which is operatively connected to every wheel motor 9 and to actuators for every hip joint 4 and every knee joint 6. These operative connections, and the constructions of the joints and wheel motors 9 themselves, are realized in such a way, that each revolute joint of the hip joint 4, the knee joint 6, and the wheel motor 9 of every wheeled leg 3 are operatable independently of each other and independently of every hip joint 4, knee joint 6, and wheel motor 9 of every other wheeled leg 3. The two non-parallel revolute joints of the hip joint 4 together with the possibility of individual positioning of the two revolute joints of the hip joint 4 and the knee joint 6 and the individual acceleration of the wheel 8 by means of its wheel motor 9 form a three-dimensional active suspension system or every wheeled leg 3.

[0031] The freedom of movement provided by the two non-parallel revolute joints of the hip joints 4 together with the possibility of individual positioning of every revolute joint of every hip joint 4 and every knee joint 6 and the individual acceleration of every wheel 8 by means of its wheel motor 9 features efficient changes of direction, switching between driving, stepping, and driving while stepping, and stabilization. In particular, it allows for omnidirectional movements: for example, stairs can be traversed by stepping, rolling, or a combination thereof, and direct movements in a direction that does not coincide with the alignment of the wheels can be achieved by stepping. Here the stabilization also includes balancing with two or more wheels 8 on the ground, e.g., in an upright position with two wheels 8 in the air.

[0032] More precisely movements can be achieved by the combination of the concepts of a) individual acceleration of each wheel 8 and b) positioning of the wheeled legs relative to the surface and relative to the torso 2. Note that b) relates to e.g., stepping movements, which can also be combined with rolling movements, and adaptions of the suspension system to change direction.

[0033] For example, changes of direction can be achieved with concept a) making use of friction and different drive torques in different locations, and with concept b) making use of mass distribution, momentum, and inertia.

[0034] The versatility of the hip joints 4 together with the possibility of individual actuation of every joint and every wheel motor 9 allow for great freedom of movement without the need of e.g., additional joints and motors to steer each wheel 8 by relocating its axis of rotation with respect to the plane p_l. Also, the versatility of the hip joints 4 is achieved by two comparably simple and robust joints instead of a fragile complex solution. These aspects moreover provide the advantage of involving minimal hardware, which in turn means lower costs and efforts, as well as less weight to be carried around. The latter fact yields less effects on the performance to be considered in physical models and less energy consumption because of the weight itself as well as because of the lowered computational power needed.

[0035] The microchip of the electronics 10 is also operatively connected to the at least one sensor and to the communication system.

[0036] The at least one sensor comprises here at least: a depth sensor and / or a camera on each side of the torso 2 and / or an Inertial Measurement Unit (IMU). Optionally, for example one or more lasers and / or one or more motor sensors for wheel motors 9 and / or one or more joint position encoders detecting the state of a joint and / or one or more actuator sensors for joints and / or one or more mass spectrometers and / or a three hundred and sixty degree camera on the top face of the torso 2 and / or a gyroscope and / or a position sensor using Global Positioning System (GPS) technology may be incorporated in or attached to the mobile robot 1. Every sensor can be incorporated in or attached to the torso 2 and / or one or more of the wheeled legs 3.

[0037] The communication system is here a combined wireless sender and receiver, preferably using Wireless Local Area Network (WLAN) technology. Other possibilities include Next Generation Mobile Networks (NGMN) technology, like e.g., 4G, 5G, Bluetooth Low Energy (BLE) technology, and Wireless Personal Area Network (WPAN) technology. The communication system can also work with more than one technology; as BLE and WPAN typically have a lower range than WLAN and NGMN, a combination of long and short range technologies can be advantageous to cover possibly simultaneous communication to external devices farther away, like a server, and devices close by, like a remote control. The communication system is preferably attached to or integrated in the torso 2.

[0038] The electronics 10 can comprise parts incorporated in the torso 2 and / or parts incorporated in one or more of the wheeled legs 3. They can comprise several instead of one microchip and possibly divide tasks among them.

[0039] A preferred use of use of the mobile robot 1 of the invention is the transport of goods and people, and most preferably goods. As shown in Fig. 1, carriers 21 for goods can for example be attached to the top face of the torso 2 of the mobile robot 1 for this purpose. Alternatively, or in addition, the torso 2 may comprise cavities and / or recesses for transportation means.

[0040] Another preferred use of the mobile robot 1 of the invention is the gathering of large-scale data by means of its at least one sensor.

[0041] Fig. 2 shows the mobile robot 1 from Fig. 1 without the carriers 21. Additionally, computer hardware 11 in the form of the laptop 11 is shown. Note that the computer hardware 11 can for example also be a general personal computer, a server, or one or more microchips incorporated in the mobile robot 1, or combinations of these options. In particular, the computer hardware 11 can also form a part of the electronics 10.

[0042] The laptop 11 communicates with the electronics 10 of the mobile robot 1 via the communication system of the latter and standard WLAN means incorporated in the laptop 11. Alternatively, if the communication system of the mobile robot 1 uses other technology, suitable means for communication therewith may be integrated in or connected to the computer hardware 11. If the computer hardware 11 is part of the electronics 10, communication to other parts of the electronics 10 is part of the electronics 10 itself.

[0043] A user interface which receives, possibly processes, and transmits input of a user to a planning software 111 on the computer hardware 11 is realized here by a graphical application on the laptop 11 and connected to the electronics 10 via the communication system.

[0044] Examples of alternative realizations of the user interface are an application on a personal mobile device, a touchscreen incorporated in the mobile robot 1, or a remote control.

[0045] The user interface can also be realized in an audial or haptic way for example.

[0046] More generally, the user interface is either incorporated in the computer hardware 11 or, at least temporarily, connected to it. In the latter case, the user interface can be incorporated in the electronics 10 of the mobile robot 1, i.e., in the electronics 10 outside of the computer hardware 11 if the computer hardware 11 forms part of the electronics 10, or it can be incorporated in or attached to the mobile robot 1 outside of the electronics 10, or it can be incorporated in or constituting a device not being part of the mobile robot 1. If the user interface is incorporated in or constitutes a separate device and the computer hardware 11 is physically separated from the electronics 10, the connection between the user interface and the computer hardware 11 can be via the electronics 10 of the mobile robot 1.

[0047] The connection between the user interface and the computer hardware 11 needs not to be continuous but can be any means to transport data from the user interface to the computer hardware 11, possibly in timeshifted steps. For example, the user interface can be a console for programming software that is then loaded once onto the computer hardware 11, or the user interface can belong to any device and / or program which is directly or through additional devices and / or programs connected to the planning software 111 via an application programming interface (API) and possibly an additional program and / or a communication system.

[0048] The communication between the computer hardware 11 and the electronics 10 of the mobile robot 1 is used for a drive control method by means of which the mobile robot 1 is operated. For this method, the planning software 111 is installed on the computer hardware 11 and, at least temporarily, connected to the user interface, i.e., here the application running on the laptop 11 comprising the user interface. To initiate the operation and / or a task of the mobile robot 1, the user can set a goal via the user interface. This goal is then handed over to the planning software 111 which receives the goal and subsequently determines a plan depending on the goal and environmental data maps. Note that alternatively, the planning software 111 may already receive the input comprising the goal via the user interface before it is loaded onto the computer hardware 11, and possibly even already while it is programmed itself; in the latter case the user interface can be the console for programming the planning software 111. In that case, the input is not received and accessed but just accessed by the planning software 111 during the execution of the method.

[0049] The goal can, for example, be a destination, e.g., a target position described by its GPS data; a compass direction; a mode, e.g., exploration, speed, or energy saving; a time, e.g., until when / how long to go in a certain direction; or combinations thereof. It can also be a more general task goal, like to keep moving and not get stuck, which can for example already be incorporated via the console as user interface when programming the planning software 111.

[0050] The data points of the environmental data maps can, for example, describe street plans, height profiles, profiles of measurement data like temperature, humidity, light, traffic, movements, or raw data from exteroceptive sensors. More generally, the environmental data maps are sets of data points associated to points on a surface or in space, which may additionally be coupled to metadata and / or timestamps. The data points may be fixed, e.g., in the case of a street map, or variable, e.g., in the case of the map reflecting repeated temperature measurements at different points, or a combination thereof. The selection of points on the surface for which data points are contained in the environmental data map may vary as well. The measurement data may be provided by external sensors, or by the at least one sensor of the mobile robot 1. External sensors may be stationary, or correspond to e.g., other mobile robots.

[0051] The plan is a set of instructions for the electronics 10 of the mobile robot 1. These instructions comprise commands on the individual actuation of each joint and each wheel motor 9, i.e., on their operation. The commands specify for example the kind of actuation and its intensity. The instructions can for example comprise commands for instantaneous execution, lists of commands to be executed successively, or commands coupled to some trigger, or combinations thereof. A trigger can, for example, be a point in time, a relative time, or a geographical position. In the case of a list of commands to be executed successively and / or commands coupled to some trigger, the execution of the instructions by the electronics 10 of the mobile robot 1 according to the plan may cause the mobile robot 1 to follow a certain trajectory.

[0052] The planning software 111 determines the set of instructions constituting the plan using a learning method that employs the concepts of changing direction and overcoming obstacles by adjusting the position of joints and changing direction by adjusting the acceleration of the wheels. More generally, reinforcement learning, learning, optimal control, and optimization methods can be employed, and motion libraries may be used. The optimization goals and constraints for the determination of the plan are given by the goal set by the user, the data from the environmental data maps, the construction details of the mobile robot 1, and possibly further available data. The latter includes information on possible payload provided either by sensors or by the user via the user interface and / or data from a digital twin. Computationally intensive tasks are most preferably performed here with the involvement of at least one neural network 112 schematically shown in Fig. 2. The at least one neural network 112 can be part of the computer hardware 11. Alternatively, the at least one neural network 112 can be implemented in an external device, like a server or server complex, which can communicate with the computer hardware 11 via, for example, WLAN. In the case of more than one neural networks, they can be implemented in the same hardware or distributed across different hardware, including computer hardware 11. If the environmental data maps are variable, the determination of the plan is either based on their state at a given point in time, or on the average over a given period of time.

[0053] After the determination of the plan, the planning software 111 initiates the transmission of the plan to the electronics 10 of the mobile robot 1 via the communication systems of the latter and the computer hardware 11.

[0054] The determination of the plan and the transmission of the plan to the electronics 10 of the mobile robot 1 are initiated preferably as soon as the necessary data is available.

[0055] Note that the user input via the user interface may comprise more information than just the goal. In that case, this information may be transmitted to and processed by the planning software 111 as well, or this information and / or the goal may be transmitted to and processed by the electronics 10, or a combination of these options may apply.

[0056] After the plan is determined and transmitted to the electronics 10 of the mobile robot 1, an execution software 101 installed on the electronics 10 of the mobile robot 1 receives the plan and subsequently starts to execute the instructions comprised by the plan.

[0057] At the same time, or starting before that, or later, e.g., after a given timespan, the execution software 101 starts to orchestrate the performance of measurements with the at least one sensor. This is either done according to predefined settings of the execution software 101 and / or the electronics 10, and / or according to instructions received from the computer hardware 11. The generated measurement data is either transmitted from the at least one sensor to the electronics 10 via its operative connection to the at least one sensor or transmitted by means of communication devices directly from the at least one sensor to the computer hardware 11. If the measurement data is transmitted to the electronics 10 it is either first processed there or sent directly to the computer hardware 11 via the communication system. There may be additional measurement data which is only processed by the electronics 10 and not sent to the computer hardware 11.

[0058] The planning software 111 on the computer hardware 11 receives the measurement data and subsequently determines adjustments to the plan based on the measurement data, and possibly further data like e.g., the plan and / or updated states of variable environmental data maps and / or data used to determine the plan and / or measurement data provided by external sensors. The determination of the adjustments is realized with an inference process. This can use a learning method which may employ the concepts of changing direction and overcoming obstacles by adjusting the position of joints and changing direction by adjusting the acceleration of the wheels 8. This learning method may be the same as the one employed for the determination of the plan. Computationally intensive tasks are preferably outsourced to at least one neural network 112, similarly to the case of the determination of the plan; if both the determination of the plan and the determination of the adjustments to the plan involve at least one neural network 112, these can be the same neural networks 112 or distinct ones. The adjustments may be a set of sequences of instructions for the electronics 10 of the mobile robot 1, each coupled with instructions on which part of the plan should be replaced by or run in parallel to it. Alternatively, the adjustments can, for example, be a complete updated version of the plan which replaces the plan or is executed in parallel or subsequent to the plan. The adjustments to the plan allow for reactions to environmental occurrences which are or were not captured in the environmental data maps, like previously unknown constitutions of parts of the surface which the mobile robot 1 must traverse, and / or which are variable, like the traffic or the humidity. Moreover, the adjustments of the plan can also be used to smoothen transitions prescribed by the plan, e.g., between phases of motion, when a sequence of instructions from a motion library was inserted, or between composed motion sequences from motion libraries.

[0059] After determination, the adjustments to the plan are transmitted, upon initiation by the planning software 111, to the electronics 10 of the mobile robot 1 via the communication systems of the latter and the computer hardware 11.

[0060] The execution software 101 then receives the adaptions to the plan and adjusts the operation of the joints and the wheel motors 9 accordingly. Note that the steps from gathering measurement data to adjusting the operation according to the adjustments of the plan implement real-time controlling and may be repeated regularly or even continuously.

[0061] The above-described transmissions of data, computations, and executions are initiated preferably as soon as the necessary data is available.

[0062] If the computer hardware 11 is part of the electronics 10, the data transfer forth and back between the planning software 111 and the execution software 101 may be implemented by hardware connections within the electronics 10. If the computer hardware 11 is part of the electronics 10, the planning software 111 and the execution software 101 can further be implemented in a single software solution, i.e., in a common software system or even within a single program.

[0063] If the planning software 111 and the execution software 101 are implemented in a single software solution, the data transfer is determined by this single software solution and individual steps of the method can be merged. For example, the determination of the plan in step a and the determination of the adjustments to the plan in step c can be done using the same algorithm implementing a continuous inference process. This algorithm initially gets as input the goal and the environmental data maps and from this computes and then outputs instantaneous commands for the joint positions and motor states. Meanwhile, the electronics 10 already gather measurement data from the at least one sensor of the mobile robot 1 and provide the measurement data to the algorithm. From the measurement data, the goal, the environmental maps which are possibly updated, and possibly additional data, the algorithm then again computes and outputs instantaneous commands for the joint positions and motor states. In that case, the determination of the commands and the execution thereof can be realized continuously and nearly instantaneously. The inference process implemented by the algorithm can be run several times per second and the commands are recalculated every time. This algorithm for example in order to achieve a certain task manages the mobile robot 1 to drive, step, make one or more of the wheeled legs 3 slip, and adjust the positions of the joints for steering; and all of this possibly at the same time and without further human guidance.

[0064] In the case of the computer hardware 11 comprising external components, like a server, as well as internal components of the mobile robot 1, like a microchip, the planning software 111 may be divided into several subroutines, where, for example, subroutines determining the plan run on the external server and subroutines determining the adjustments to the plan run on the internal microchip.

[0065] There can be more input from the user via the user interface also after the initial input comprising the goal. This can also include a replacement of the goal. For example, the mobile robot 1 may receive input via a remote control.

[0066] The measurement data, the plan, the adjustments to the plan, particular decisions made while the determination of the plan or the adjustments to the plan, and possibly other data may be gathered, possibly processed, and used for future determinations of plans and adjustments to plans. If there are more than one mobile robot 1 which can provide such data, those mobile robots 1 may share their data among them.

[0067] Fig. 3 shows a top view on the mobile robot 1 from Figures 1 and 2. In particular, the position of the hip joint 4 of the wheeled leg 3 in stepping position compared to the position of the hip joints 4 of the wheeled legs 3 not in stepping position is shown.

[0068] As shown in Figures 1 to 3, the mobile robots 1 torso 2 comprises here a shell housing the electronics 10 and possibly fragile hardware components and protecting them from external influences. There can be cavities in the shell or in the torso 2 underneath the shell to transport goods. Preferably, the shell has an aerodynamic shape.

[0069] The form of the upper leg 5 of each wheeled leg 3 is here cylindric with a thickened end towards the hip joint 4. The form of the lower leg 7 of each wheeled leg 3 is here longitudinally S-shaped and has a rectangular cross section. Alternatives include upper legs 5 and lower legs 7 being of the same shape but possibly of different size.

[0070] The mobile robot 1 comprises here light sources at the front end of the torso 2.

[0071] The mobile robot 1 presented here is capable of omnidirectional movement on any terrain without the use of steerable wheels or a steering mechanism at the wheels. Its minimal hardware design allows for efficient flat terrain and rough terrain traversal, versatility over obstacles like steps and stairs, and payload capabilities for sensors and goods in any indoor and outdoor space. In particular, the joint architectures and the three-dimensional active suspensions of the wheeled legs 3 allow for great freedom of movement. They allow to control the position of the wheels 8 in the cartesian x, y and z direction. The omnidirectional movement of the mobile robot 1 is achieved by deploying an algorithm that adjusts the positions of the joints and the speed of the wheels 8 in relation to one another, and by utilizing one or more of the following steering concepts: violating friction constraints by adapting ground reaction forces through the simultaneous positioning of the center of mass and wheel speed adaptation; hybrid mobility by stepping and rolling at the same time; and direction changes by adapting the suspension system, i.e., rotating the robot's torso around the yaw. Learning algorithms exploit the hardware's design and automatically find the appropriate steering concepts, and real-time controlling robustifies the performance against the variability of the physical world and narrows down the simulation-to-reality gap if training is realized with simulations in addition to or instead of with a real mobile robot 1.

[0072] This allows to seamlessly track rapidly changing velocity commands in any direction while other systems need to switch from feet to wheels or rotate the steering motors to satisfy the non-holonomic constraints of the system. This also allows the users to provide any input about how and when to transition from one steering strategy to another.

[0073] The here presented approach with the use of neural networks for accomplishing (a) automatic steering motions with non-steerable wheels while (b) also overcoming challenging terrains like steps and stairs is the only one to achieve omnidirectional motions of the mobile robot 1 in real-time, i.e., on high update rates over 5 Hz.

[0074] The learning method achieves (real-time) behaviours for the non- steerable platform. These behaviours comprise steerable behaviour over any terrain. The solution presented here moreover is the minimal solution to achieve efficient motions over flat and difficult terrain. Neural networks need to be deployed to achieve this since other known methods cannot achieve this and are slow in computation.

[0075] The achieved motions are e.g. : i) violation of friction constraints by adapting ground reaction forces through the simultaneous center of mass and wheel speed adaptation, ii) hybrid mobility by stepping and rolling at the same time, and iii) direction changes by adapting the suspension system, i.e., rotating the robot's torso around yaw.

[0076] These motions are discovered i) without heuristics engineering, ii) without data from other animals since such a morphology does not exist in nature, and iii) the motions have to be self-discovered through learning algorithms.

[0077] Our neural network, differently from other proposed, can simultaneously take care of multiple functions: balancing the robot, steering, stepping, crossing obstacles, and tracking velocity commands.

[0078] Regarding the use of the mobile robot 1 for transport, the ability of the drive control method to include the payload in its planning from the beginning, and then to react to possible movements of the transported goods or people during the ride by means of real-time controlling ensures the robust and efficient execution of tasks. Moreover, the minimal hardware solution provides not only less manufacturing steps, but also the advantages of less energy consumption and less net weight, which on the one hand allows for more payload and on the other hand also for more sensors to ensure the movement of the mobile robot 1 is safe for itself, its payload, and the other traffic members.

[0079] Note that the statements made in this section also apply mutatis mutandis in the case of one or more features exhibiting properties alternative to those described in this section.

[0080] LIST OF REFERENCE NUMERALS

[0081] 1 Mobile robot

[0082] 2 Torso

[0083] 21 Carrier

[0084] 22 Light source

[0085] 3 Wheeled leg

[0086] 4 Hip joint

[0087] 41 Hip joint component

[0088] 5 Upper leg

[0089] 6 Knee joint

[0090] 7 Lower leg

[0091] 8 (Non-steerable) wheel

[0092] 9 Wheel motor

[0093] 10 Electronics

[0094] 101 Execution software

[0095] 11 Computer hardware

[0096] 111 Planning software

[0097] 112 Neural network

Claims

PATENT CLAIMS1. Drive control method for a mobile robot (1) comprising a torso(2) and at least four wheeled legs (3), with each wheeled leg (3) comprising a hip joint (4), an upper leg (5), a knee joint (6), a lower leg (7), and a non-steerable wheel (8) with an attached wheel motor (9); wherein the mobile robot (1) is equipped with at least one energy source, at least one sensor, a communication system, and electronics (10) which are operatively connected to each of the former as well as to the hip joint (4), the knee joint (6), and the wheel motor (9) of every wheeled leg (3); wherein the hip joint (4) of every wheeled leg(3) comprises two revolute joints with mutually non-parallel axes; and each revolute joint of the hip joint (4), the knee joint (6), and the wheel motor (9) of every wheeled leg (3) are operatable by means of the electronics (10) independently of each other and independently of every hip joint (4), knee joint (6), and wheel motor (9) of every other wheeled leg (3); the method comprising the following steps: a. by means of a planning software (111) on a computer hardware (11) forming part of the electronics (10) or being physically separated from the electronics (10):- accessing input provided via a user interface and comprising a goal, and- determining a plan depending on the goal and environmental data maps, b. by means of an execution software (101) on the electronics (10) of the mobile robot (1):- according to the plan, operating the hip joint (4), the knee joint (6), and the wheel motor (9) of every wheeled leg (3); c. real-time controlling:- by means of the execution software (101): performing measurements with the at least one sensor,- by means of the planning software (111): determining adjustments to the plan based on the measurement data,- by means of the execution software (101): according to the plan and the received adjustments to the plan, operating the hip joint (4), the knee joint (6), and the wheel motor (9) of every wheeled leg (3); wherein the determination of the plan in step a and / or the determination of the adjustments of the plan in step c is implemented using a learning method that employs the concepts of changing direction by adjusting the position of joints and changing direction by adjusting the acceleration of the nonsteerable wheels (8), and where the learning method is performed with the involvement of at least one neural network (112).

2. Method according to claim 1, wherein the determination of the plan in step a and the determination of the adjustments of the plan in step c is implemented using a learning method that employs the concepts of changing direction by adjusting the position of joints and changing direction by adjusting the acceleration of the non-steerable wheels (8), and where the learning method is performed with the involvement of at least one neural network (112).

3. Method according to one of the preceding claims, wherein the planning software (111) and the execution software (101) are implemented in a single software solution.

4. Method according to claim 3, wherein one or more individual steps of the method are merged to one or more combined steps.

5. Method according to one of the preceding claims, wherein by means of the execution software (101), the performance of measurements with the at least one sensor are orchestrated.

6. Method according to one of the preceding claims, wherein determining adjustments to the plan based on the measurement data by means of the planning software (111) in step c is realized with an inference process.

7. Use of at least one neural network (112) to implement a drive control method for a mobile robot (1), wherein the at least one neural network (112) is integrated in a computer hardware (11) and learning algorithms running on the at least one neural network (112) are employed by a planning software (111) running on the computer hardware (11) and / or an execution software (101) running on electronics (10) of the mobile robot (1) which optimize, with respect to a predefined target, operating the speed of non-steerable wheels (8) and the position of hip joints (4) and knee joints (6) independently in a multiplicity of wheeled legs (3) of the mobile robot (1).

8. Use of at least one neural network (112) according to claim 7, executed in the following detailed steps: a. by means of the planning software (111) on the computer hardware (11) forming part of the electronics (10) or being physically separated from the electronics (10):- accessing input provided via a user interface and comprising a goal, and- determining a plan depending on the goal and environmental data maps,b. by means of the execution software (101) on the electronics (10) of the mobile robot (1):- according to the plan, operating the hip joint (4), the knee joint (6), and a wheel motor (9) of every wheeled leg (3); c. real-time controlling:- by means of the execution software (101): performing measurements with the at least one sensor,- by means of the planning software (111): determining adjustments to the plan based on the measurement data,- by means of the execution software (101): according to the plan and the received adjustments to the plan, operating the hip joint (4), the knee joint (6), and the wheel motor (9) of every wheeled leg (3); wherein the determination of the plan in step a and / or the determination of the adjustments of the plan in step c is implemented using a learning method that employs the concepts of changing direction by adjusting the position of joints and changing direction by adjusting the acceleration of the nonsteerable wheels (8), and where the learning method is performed with the involvement of the at least one neural network (112).

9. Mobile robot (1) to be operated by a drive control method according to one of the claims 1-6 and / or by means of the use of at least one neural network (112) according to one of the claims 7-8; the mobile robot (1) comprising a torso (2) and at least four wheeled legs (3), with each wheeled leg (3) comprising a hip joint (4), an upper leg (5), a knee joint (6), a lower leg (7), and a non-steerable wheel (8) with an attached wheel motor (9); wherein the mobile robot (1) is equipped withat least one energy source, at least one sensor, a communication system, and electronics (10) which are operatively connected to each of the former as well as to the hip joint (4), the knee joint (6), and the wheel motor (9) of every wheeled leg (3); characterized in that the hip joint (4) of every wheeled leg (3) comprises two revolute joints with mutually non-parallel axes; and each revolute joint of the hip joint (4), the knee joint (6), and the wheel motor (9) of every wheeled leg (3) are operatable by means of the electronics (10) independently of each other and independently of every hip joint (4), knee joint (6), and wheel motor (9) of every other wheeled leg (3).

10. Mobile robot (1) according to claim 9, wherein the hip joint (4) of every wheeled leg (3) is realized by means of a hip joint component (41) of the form of a uniform prism with a first and a second connection, where the first connection protrudes centrally from and perpendicular to one of its bases, is on one end fixed to an elongated side of the upper leg (5) of the wheeled leg (3) and is on the opposite end with respect to its central longitudinal axis rotatably attached to the hip joint component (41), and where the second connection protrudes from the hip joint component (41) along a central axis parallel to its bases, is on one end fixed to the latter and is on the opposite end with respect to its central longitudinal axis rotatably attached to the torso (2).

11. Mobile robot (1) according to one of the claims 9-10, wherein the hip joint (4), the knee joint (6), and the wheel motor (9) of every wheeled leg (3) are each actuated either electrically, hydraulically, pneumatically, or by a combination of these options.

12. Mobile robot (1) according to one of the claims 9-11, wherein the at least one sensor comprises at least: at least one depth sensor and / or at least one camera and / or at least one Inertial Measurement Unit (IMU).

13. Mobile robot (1) according to claim 12, wherein the at least one sensor comprises: one or more lasers and / or one or more motor sensors for wheel motors (9) and / or one or more joint position encoders and / or one or more actuator sensors for joints and / or one or more mass spectrometers.

Images