WO2018154153A2 - Method for designing modular robots - Google Patents

Method for designing modular robots Download PDF

Info

Publication number
WO2018154153A2
WO2018154153A2 PCT/ES2017/070781 ES2017070781W WO2018154153A2 WO 2018154153 A2 WO2018154153 A2 WO 2018154153A2 ES 2017070781 W ES2017070781 W ES 2017070781W WO 2018154153 A2 WO2018154153 A2 WO 2018154153A2
Authority
WO
WIPO (PCT)
Prior art keywords
robot
modules
model
modular
task
Prior art date
Application number
PCT/ES2017/070781
Other languages
Spanish (es)
French (fr)
Other versions
WO2018154153A3 (en
Inventor
Victor MAYORAL VILCHES
Alejandro HERNANDEZ CORDERO
Risto KOJCEV
Irati ZAMALLOA UGARTE
Original Assignee
Erle Robotics, S.L.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Erle Robotics, S.L. filed Critical Erle Robotics, S.L.
Priority to PCT/ES2017/070781 priority Critical patent/WO2018154153A2/en
Publication of WO2018154153A2 publication Critical patent/WO2018154153A2/en
Publication of WO2018154153A3 publication Critical patent/WO2018154153A3/en

Links

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/16—Programme controls
    • B25J9/1602—Programme controls characterised by the control system, structure, architecture
    • B25J9/161—Hardware, e.g. neural networks, fuzzy logic, interfaces, processor
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/08—Programme-controlled manipulators characterised by modular constructions
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/16—Programme controls
    • B25J9/1602—Programme controls characterised by the control system, structure, architecture
    • B25J9/1605—Simulation of manipulator lay-out, design, modelling of manipulator
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/008—Artificial life, i.e. computing arrangements simulating life based on physical entities controlled by simulated intelligence so as to replicate intelligent life forms, e.g. based on robots replicating pets or humans in their appearance or behaviour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/39—Robotics, robotics to robotics hand
    • G05B2219/39064—Learn kinematics by ann mapping, map spatial directions to joint rotations
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/40—Robotics, robotics mapping to robotics vision
    • G05B2219/40304—Modular structure
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/40—Robotics, robotics mapping to robotics vision
    • G05B2219/40499—Reinforcement learning algorithm

Definitions

  • the object of the invention is framed in the technical field of robotics.
  • the object of the invention is oriented to activities such as the configuration or training of modular robots using neuromorphic artificial intelligence techniques to perform one or more tasks.
  • integration-oriented consists of building robots through continuous integration of non-interoperable modules. This impacts the flexibility and reconfiguration capacity of these robots for other purposes. In opposition to this, modular robots promise interoperability and ease of retrofitting.
  • Robot control project Traditional methods for programming and configuring robots involve a structured process typically known as the "robotic control project". This process requires a fine adjustment of each of the logical blocks involved in the process, from the estimation of the state, to the low level control of the robots. This project requires deep experience to obtain accurate results and is typically error sensitive. The mistakes spread throughout the project and any change in the modules or the physical structure of the robot requires relevant modifications to the project.
  • US20150258683A1 defines generic modular robotic devices with embedded artificial intelligence based on deep learning techniques and trained to execute a particular given task.
  • Said device includes the option of adding a camera, inertial sensors, audio input and the production of an output signal, however the document does not define a clear concept of modularity in terms of robot modules or these are used for the processes of training.
  • the robot apparatus defined in the document addresses a single task and there is no explicit mention of the internal model of the robot that is used by the computational brain for purposes of reconfiguration.
  • each "removable enclosure" could include its own reconfiguration means through additional detection, or processing itself.
  • document US20140019392A1 proposes a concept of extensible robot device defined as a mechanical "neuromorphic device" with embedded artificial intelligence.
  • the patent provides an interesting definition of how embedded artificial intelligence allows a user to train a robotic device in a concept similar to the way in which training of a domesticated animal is performed as it is used with a dog or cat (for example , a positive / negative feedback training paradigm).
  • This behavioral control structure is based on the concept of Artificial Neural Networks (ANN), which simulates the neuronal / synaptic activity of the brain of a living organism.
  • ANN Artificial Neural Networks
  • the documentation does not specify characteristics for each of the different specialized neuromorphic devices next to the brain (for example, sensors, actuators, etc.) that could influence the overall behavior and learning process. Additionally, the document does not consistently consider non-neuromorphic robot modules and their inclusion in the robot's overall behavior. In addition to this, the content does not include the robot's internal model and its physically available modules as a critical input to neuromorphic device devices for proper configuration.
  • the object of the invention described herein provides a method for the configuration or training of robots constructed from modular modules using artificial intelligence techniques for the execution of one or more various tasks
  • Modular robots require methods for their modules that incorporate a highly modular and interchangeable software and hardware architecture. With this, the modular robot modules, packaged in a module that typically contains software and hardware, could interrelate with multiple robotic bodies for a variety of applications.
  • the objective of the present invention is to describe a method for programming or training robots constructed from modular parts for the execution of one or more tasks using techniques that mimic the neurobiological architectures present in the nervous system. These techniques typically make use of a combination of Artificial Neural Networks — ANN— and Learning by Reinforcement — RL — techniques to represent and acquire the desired intelligence by creating models that can be dynamically retrained depending on the available and operational parts in the modular robot. Finally, the present invention empowers the creation of modular robots whose behavior is generalized, if physically feasible, regardless of the parts available for their activity.
  • Module Part or element of something more complex. It could represent only hardware, software or a combination of both.
  • a module that typically integrates both hardware and software and special features that provide means to facilitate its design, interoperability, easy reconfiguration and reuse.
  • the modules belong to one of the following classes: detection that means perceiving and detecting the environment, acting, which allows robots to produce a physical change in the environment, communication, which provides means of interconnection either between modules or with external systems , processing, to perform most of the computationally expensive tasks within the robot, power that provide global power to the modular robot, or user interface (Ul) that provides means to interact with the robot modular.
  • Robot model Representation of the modules inside the robot including but not limited to their physical characteristics or type.
  • the object of the invention provides means for the construction of an "M” model preferably composed of Artificial Neural Networks, programming logic or a combination of both.
  • Said model “M” is configured to receive a certain number of inputs "I” and the robot model “RM”, the model “M” is configured to generate a certain number of outputs "O" for the respective actuators of the robot " R “modeled by the” RM “robot model assuming that the” R “robot can perform a certain” T "task, specifically determining that it is physically feasible, regardless of the number of" N "modules of which the "R” robot.
  • the "M” model should be updated (or retrained) whenever a change in the "RM” robot model takes place (which could be activated by the dynamic removal or addition of robot modules or modules " N ").
  • the update of the "M” model can be carried out through reinforcement learning, which can be understood in a conceptually similar way to the way in which the training of a domesticated animal such as a dog or cat is carried out (for example, a negative positive feedback training paradigm).
  • the object of the invention conceives a possible combination of transfer learning and reinforcement learning allowing the "M” model to adapt its behavior when a new module is connected while maintaining the previous knowledge acquired. Finally, this implies that the "M” model is considered to be able to perform the "T” task under a variable number "N" of modules.
  • the object of the invention allows the training of modular robots to achieve a given task under different hardware configurations while maintaining the same performance.
  • the invention defines an "M” model composed of Artificial Neural Networks, programmed logic or a combination of both that makes use of a dynamic description of the "N" modules within the "R” robot (specifically the robot model “RM”) to dynamically adapt its behavior.
  • Figure 1.- Describes a flow chart representing a given "M" model of a preferred embodiment of the method object of the invention.
  • Figure 2. Shows a 3D representation of a modular robot comprising three joints and five modules, in which no power and communication details have been represented.
  • Figure 3. Shows a 3D representation of a modular robot comprising five modules, in which the objective task (T) is also represented.
  • Figure 4. Shows a 3D representation that a modular robot comprising five modules and three joints in which the last joint is eliminated (resulting in four modules), in which the objective task (T) is also represented.
  • Figure 5. Shows a 3D representation of the modular robot of figure 4 after the last joint has been removed.
  • a method for configuring a modular robot (R) - hereinafter robot (R) - comprising a series of modules (N) or modules, defined by actuators (A1, A2, A3 ).
  • Each module (N) could include its own calculation means and / or additional detection means, at least said modules (N) being provided with a processing unit so that orders and instructions can be processed.
  • a model (M) is composed of Artificial Neural Networks, programming logic or a combination of both and is adapted to receive a number of inputs (I) and a robot model (RM) which is a representation of the modules (N) that exist at a given time in the robot (R).
  • Model (M) generates a certain number of outputs (O) for one or more robot actuators (R) so that the robot (R) performs a given task (T) regardless of the number of modules (N) in the robot, if it is physically feasible.
  • the robot model (RM) may be a list describing each of the modules (N) of the robot (R) and how they relate to each other; said modules (N) being modeled as modeled modules ( ⁇ ').
  • the inputs (I) at least one of the modeled modules ( ⁇ ') of the robot (R), specifically the modeled actuators ( ⁇ 1', ⁇ 2 ', A3') react to the input (I) mentioned.
  • the input (I) may comprise information related to the execution of a task (T) given by the robot and / or sensory information relating to one or more of an actuator ( ⁇ 1 ', ⁇ 2', ⁇ 3 ').
  • the input (I) can include information related to the execution of a given task (T), so that the robot (R) carries out the task (T).
  • the modules (N) To carry out the task (T), the modules (N) must execute a series of orders and actions accordingly; Therefore, in the real world, the actuators (A1, A2, A3) of the robot (R) should execute some orders and activate their respective movements so that the robot (R) performs the task (T).
  • Corresponding A3 ' are used to evaluate the possible combinations of actions required to perform the task (T).
  • the model (M) processes the inputs (I) and generates the respective outputs (O) from the modeled actuators ( ⁇ 1 ', ⁇ 2', ⁇ 3 ') - not shown in the figures - of the robot model ( RM) so that it performs the given task (T) regardless of the number of modules (N) in the robot model (RM), if it is physically feasible according to either the robot model (RM) or the robot (R) itself. Note that the number of modules (N) inside the robot is also represented in the robot model (RM).
  • the robot (R) has a terminal effector (EF) at its end with a laser pointer towards the workspace.
  • the objective, task (T) of the robot (R) is for the terminal effector (EF) to reach a certain point in the space - marked as "objective” in Figure 3 - in the workspace by generating movements in the actuators (A1, A2, A3), while being coordinated by the brain (B), which contains the model (M) comprising the instructions related to the behavior required for each of the actuators (A1, A2, A3) so that the terminal effector (EF) reaches the certain point in the space — marked as “objective” in Figure 3—.
  • the method of the invention allows the robot (R) provided with the third actuator (A3) to adopt the model (M) so that the robot (R) achieves the same "target” and vice versa.
  • the method of the invention comprises performing a process for transforming the robot model (RM) in a way that allows to represent categorical characteristics of the robot model (RM) in a format that behaves more efficiently with classification and regression algorithms typically used in reinforcement learning.
  • the result is called the Encoded Robot Model (abbreviated "ERM").
  • the method provided herein comprises the finding of a fixed-size coding for the ERM (for example: in this particular embodiment using the Oblivion-Fixed-Fixed-Size Ordinary-Forgetting Encoding method) called Fixed Size Coded Robot Model (abbreviated "FERM”) to train or adapt the model (M) through reinforcement learning techniques using a combination of inputs (I) and FERM.
  • FERM Fixed Size Coded Robot Model
  • the model (M) could be adapted when new modules (N) are added (or removed) provided there is a physical feasibility to complete the task (T) with a robot (R) that comprises said modules (N), specifically reaching the "objective.” This is achieved by the transfer of the output (O) to the modules (N) of the robot (R) as orders to be executed to perform the task (T).

Abstract

A method is provided for designing modular robots comprising a certain number of modules. The method according to the present invention allows the modular robot to carry out a given task, whatever the modules that define the robot, whenever this is physically possible. This means that a certain robot comprising a certain number of modules can carry out the task, and when a new module is added to the design, the method according to the invention allows the redesigned robot to carry out the given task. The same is true when a module is removed.

Description

MÉTODO PARA CONFIGURAR ROBOTS MODULARES  METHOD FOR CONFIGURING MODULAR ROBOTS
OBJETO DE LA INVENCIÓN OBJECT OF THE INVENTION
El objeto de la invención se enmarca en el campo técnico de la robótica. The object of the invention is framed in the technical field of robotics.
Más específicamente, el objeto de la invención está orientado a actividades tales como la configuración o entrenamiento de robots modulares usando técnicas de inteligencia artificial neuromórfica para realizar una o más tareas. More specifically, the object of the invention is oriented to activities such as the configuration or training of modular robots using neuromorphic artificial intelligence techniques to perform one or more tasks.
ANTECEDENTES DE LA INVENCIÓN BACKGROUND OF THE INVENTION
El planteamiento tradicional para la construcción de robots, el planteamiento The traditional approach to building robots, the approach
"orientado a integración", consiste en construir robots mediante integración de modo continuo de módulos no interoperativos. Esto impacta en la flexibilidad y la capacidad de reconfiguración de dichos robots para otras finalidades. En oposición a esto, los robots modulares prometen interoperabilidad y facilidad de readaptación. "integration-oriented", consists of building robots through continuous integration of non-interoperable modules. This impacts the flexibility and reconfiguration capacity of these robots for other purposes. In opposition to this, modular robots promise interoperability and ease of retrofitting.
Los métodos tradicionales para programación y configuración de robots implican un proceso estructurado típicamente conocido como el "proyecto de control robótico". Dicho proceso requiere un ajuste fino de cada uno de los bloques lógicos implicados en el proceso, desde la estimación del estado, hasta el control de bajo nivel de los robots. Este proyecto requiere una profunda experiencia para obtener resultados precisos y es típicamente sensible a errores. Las equivocaciones se propagan a todo lo largo del proyecto y cualquier cambio en los módulos o la estructura física del robot requiere modificaciones relevantes en el proyecto.  Traditional methods for programming and configuring robots involve a structured process typically known as the "robotic control project". This process requires a fine adjustment of each of the logical blocks involved in the process, from the estimation of the state, to the low level control of the robots. This project requires deep experience to obtain accurate results and is typically error sensitive. The mistakes spread throughout the project and any change in the modules or the physical structure of the robot requires relevant modifications to the project.
Dichas limitaciones se aplican intensamente a los robots modulares por lo que se deberían considerar nuevas técnicas para configurar, programar o entrenar robots modulares de modo efectivo.  These limitations apply intensely to modular robots, so new techniques should be considered to configure, program or train modular robots effectively.
Más relacionados con los métodos que embeben tecnologías neuromórficas, el documento US20150258683A1 define unos dispositivos robóticos modulares genéricos con inteligencia artificial embebida basándose en técnicas de aprendizaje profundo y entrenados para ejecutar una tarea dada en particular. Dicho dispositivo incluye la opción de añadir una cámara, sensores inerciales, entrada de audio y la producción de una señal de salida, sin embargo el documento no define un concepto claro de modularidad en términos de módulos del robot o estos se usan para los procesos de entrenamiento. De modo similar, el aparato robot definido en el documento se dirige a una única tarea y no hay mención explícita al modelo interno del robot que se usa por el cerebro computacional con finalidades de reconfiguración. Además, la patente no menciona ni considera que cada "recinto extraíble" podría incluir sus propios medios de reconfiguración a través de detección adicional, ni de procesamiento por sí mismo. More related to the methods that embed neuromorphic technologies, US20150258683A1 defines generic modular robotic devices with embedded artificial intelligence based on deep learning techniques and trained to execute a particular given task. Said device includes the option of adding a camera, inertial sensors, audio input and the production of an output signal, however the document does not define a clear concept of modularity in terms of robot modules or these are used for the processes of training. Similarly, the robot apparatus defined in the document addresses a single task and there is no explicit mention of the internal model of the robot that is used by the computational brain for purposes of reconfiguration. In addition, the patent does not mention or consider that each "removable enclosure" could include its own reconfiguration means through additional detection, or processing itself.
Además de los documentos anteriores, el documento US20140019392A1 propone un concepto de dispositivo robot extensible definido como un "aparato neuromórfico" mecánico con inteligencia artificial embebida. La patente proporciona una definición interesante sobre cómo la inteligencia artificial embebida permite a un usuario entrenar un dispositivo robótico de una manera conceptualmente similar al modo en el que se realiza el entrenamiento de un animal domesticado tal como se usa con un perro o gato (por ejemplo, un paradigma de entrenamiento de realimentación positiva/negativa). Esta estructura de control conductual se basa en el concepto de Redes Neuronales Artificiales (ANN), que simula la actividad neuronal/sináptica del cerebro de un organismo vivo. Estas técnicas son bien conocidas dentro de una subárea de Aprendizaje de Máquina denominada Aprendizaje por Refuerzo y que se basa típicamente en actuaciones biológicamente inspiradas tales como las ANN. La documentación sin embargo no especifica características para cada uno de los dispositivos neuromórficos especializados diferentes junto al cerebro (por ejemplo, sensores, actuadores, etc.) que podrían influir en el comportamiento global y en el proceso de aprendizaje. Adicionalmente, el documento no considera consistentemente módulos de robot no neuromórficos y su inclusión en el comportamiento global del robot. Además de esto, el contenido no incluye el modelo interno del robot y sus módulos físicamente disponibles como una entrada crítica a los dispositivos de aparatos neuromórficos para su apropiada configuración.  In addition to the above documents, document US20140019392A1 proposes a concept of extensible robot device defined as a mechanical "neuromorphic device" with embedded artificial intelligence. The patent provides an interesting definition of how embedded artificial intelligence allows a user to train a robotic device in a concept similar to the way in which training of a domesticated animal is performed as it is used with a dog or cat (for example , a positive / negative feedback training paradigm). This behavioral control structure is based on the concept of Artificial Neural Networks (ANN), which simulates the neuronal / synaptic activity of the brain of a living organism. These techniques are well known within a sub-area of Machine Learning called Reinforcement Learning and is typically based on biologically inspired actions such as ANN. The documentation however does not specify characteristics for each of the different specialized neuromorphic devices next to the brain (for example, sensors, actuators, etc.) that could influence the overall behavior and learning process. Additionally, the document does not consistently consider non-neuromorphic robot modules and their inclusion in the robot's overall behavior. In addition to this, the content does not include the robot's internal model and its physically available modules as a critical input to neuromorphic device devices for proper configuration.
El uso de módulos inteligentes que embeben capacidades de detección en el robot es un aspecto tratado por los documentos US6995536 y DE4A25112 en donde se usan sensores inerciales para estimar la posición del efector terminal en un robot. También usando sensores, el documento PCT/ES2017/070437 describe la adaptación y autoconfiguración de módulos de robots para su uso en diferentes robots así como la creación de un modelo global y dinámico del robot sin conocimiento previo de su estructura.  The use of intelligent modules that embed detection capabilities in the robot is an aspect treated by US6995536 and DE4A25112 where inertial sensors are used to estimate the position of the terminal effector in a robot. Also using sensors, document PCT / ES2017 / 070437 describes the adaptation and self-configuration of robot modules for use in different robots as well as the creation of a global and dynamic model of the robot without prior knowledge of its structure.
DESCRIPCIÓN DE LA INVENCIÓN DESCRIPTION OF THE INVENTION
El objeto de la invención descrito por el presente proporciona un método para la configuración o entrenamiento de robots construidos a partir de módulos modulares usando técnicas de inteligencia artificial para la ejecución de una o varias tareas. The object of the invention described herein provides a method for the configuration or training of robots constructed from modular modules using artificial intelligence techniques for the execution of one or more various tasks
Los robots modulares requieren métodos para sus módulos que incorporan una arquitectura de software y hardware altamente modular e intercambiable. Con esto, los módulos de robots modulares, empaquetados en un módulo que contiene típicamente software y hardware, podrían interrelacionarse con múltiples cuerpos robóticos para una variedad de aplicaciones.  Modular robots require methods for their modules that incorporate a highly modular and interchangeable software and hardware architecture. With this, the modular robot modules, packaged in a module that typically contains software and hardware, could interrelate with multiple robotic bodies for a variety of applications.
Dado un robot modular que comprenda módulos que faciliten la detección, actuación, comunicación, procesamiento (o cognición), alimentación o interfaz con usuario. Dado que cada uno de estos módulos podría o no incluir sus propias capacidades de cálculo a través del uso de microcontroladores, microprocesadores, o cualquier otro medio de cálculo. Dado que cada módulo podría o no incluir detección adicional. El objetivo de la presente invención es describir un método para la programación o entrenamiento de robots construidos a partir de partes modulares para la ejecución de una o varias tareas usando técnicas que imiten las arquitecturas neurobiológicas presentes en el sistema nervioso. Estas técnicas hacen uso típicamente de una combinación de técnicas de Redes Neuronales Artificiales— ANN— y Aprendizaje por Refuerzo— RL— para representar y adquirir la inteligencia deseada creando modelos que puedan reentrenarse dinámicamente dependiendo de las partes disponibles y operacionales en el robot modular. Finalmente, la presente invención faculta la creación de robots modulares cuyo comportamiento se generaliza, si es físicamente factible, independientemente de las partes disponibles para su actividad.  Given a modular robot that includes modules that facilitate detection, performance, communication, processing (or cognition), power or user interface. Since each of these modules may or may not include its own calculation capabilities through the use of microcontrollers, microprocessors, or any other means of calculation. Since each module may or may not include additional detection. The objective of the present invention is to describe a method for programming or training robots constructed from modular parts for the execution of one or more tasks using techniques that mimic the neurobiological architectures present in the nervous system. These techniques typically make use of a combination of Artificial Neural Networks — ANN— and Learning by Reinforcement — RL — techniques to represent and acquire the desired intelligence by creating models that can be dynamically retrained depending on the available and operational parts in the modular robot. Finally, the present invention empowers the creation of modular robots whose behavior is generalized, if physically feasible, regardless of the parts available for their activity.
La invención se comprende mejor usando la terminología siguiente:  The invention is better understood using the following terminology:
• Módulo: Parte o elemento de algo más complejo. Podría representar únicamente hardware, software o una combinación de ambos. Un módulo que integra típicamente tanto hardware como software y características especiales que proporcionen medios para facilitar su diseño, interoperabilidad, fácil reconfiguración y reutilización. Los módulos pertenecen a una de las siguientes clases: detección que significa percibir y detectar el entorno, actuación, que permite a los robots producir un cambio físico en el entorno, comunicación, que proporciona medios de interconexión o bien entre módulos o bien con sistemas externos, procesamiento, para realizar la mayor parte de las tareas costosas computacionalmente dentro del robot, alimentación que proporcionan energía global al robot modular, o interfaz de usuario (Ul) que proporciona medios para interactuar con el robot modular. • Module: Part or element of something more complex. It could represent only hardware, software or a combination of both. A module that typically integrates both hardware and software and special features that provide means to facilitate its design, interoperability, easy reconfiguration and reuse. The modules belong to one of the following classes: detection that means perceiving and detecting the environment, acting, which allows robots to produce a physical change in the environment, communication, which provides means of interconnection either between modules or with external systems , processing, to perform most of the computationally expensive tasks within the robot, power that provide global power to the modular robot, or user interface (Ul) that provides means to interact with the robot modular.
• Modelo de robot: Representación de los módulos dentro del robot incluyendo pero sin limitarse a sus características físicas o su tipo.  • Robot model: Representation of the modules inside the robot including but not limited to their physical characteristics or type.
Suponiendo que un robot modular "R" compuesto por un número de módulos "N" potencialmente diferentes entre sí. Dada una tarea "T" a ser llevada a cabo y una representación del modelo del robot "RM", el objeto de la invención proporciona medios para la construcción de un modelo "M" compuesto preferentemente por Redes Neuronales Artificiales, lógica de programación o una combinación de ambas. Dicho modelo "M" se configura para recibir un cierto número de entradas "I" y el modelo del robot "RM", el modelo "M" se configura para generar un cierto número de salidas "O" para los actuadores respectivos del robot "R" modelizado por el modelo de robot "RM" suponiendo que el robot "R" puede llevar a cabo una cierta tarea "T", concretamente determinando que es físicamente factible, independientemente del número de módulos "N" de los que se compone el robot "R".  Assuming a modular robot "R" composed of a number of modules "N" potentially different from each other. Given a "T" task to be carried out and a representation of the "RM" robot model, the object of the invention provides means for the construction of an "M" model preferably composed of Artificial Neural Networks, programming logic or a combination of both. Said model "M" is configured to receive a certain number of inputs "I" and the robot model "RM", the model "M" is configured to generate a certain number of outputs "O" for the respective actuators of the robot " R "modeled by the" RM "robot model assuming that the" R "robot can perform a certain" T "task, specifically determining that it is physically feasible, regardless of the number of" N "modules of which the "R" robot.
Estarán presentes dentro de los módulos "N", actuadores, tales como componentes o módulos. Suponiendo que es físicamente factible, el método de la invención permite al robot "R" llevar a cabo la tarea "T" usando un cierto número de diferentes combinaciones de actuadores únicamente mediante la adaptación del modelo "M".  They will be present within the "N" modules, actuators, such as components or modules. Assuming that it is physically feasible, the method of the invention allows the robot "R" to carry out the "T" task using a certain number of different actuator combinations only by adapting the "M" model.
Para llevar a cabo esto, el modelo "M" debería actualizarse (o reentrenarse) cada vez que tiene lugar un cambio en el modelo de robot "RM" (lo que podría activarse por la eliminación o adición dinámica de módulos de robot o módulos "N").  To accomplish this, the "M" model should be updated (or retrained) whenever a change in the "RM" robot model takes place (which could be activated by the dynamic removal or addition of robot modules or modules " N ").
La actualización del modelo "M" puede llevarse a cabo a través de aprendizaje por refuerzo, que puede entenderse en una forma conceptualmente similar al modo en el que se realiza el entrenamiento de un animal domesticado tal como un perro o gato (por ejemplo, un paradigma de entrenamiento de realimentación positiva negativa).  The update of the "M" model can be carried out through reinforcement learning, which can be understood in a conceptually similar way to the way in which the training of a domesticated animal such as a dog or cat is carried out (for example, a negative positive feedback training paradigm).
El objeto de la invención concibe una combinación posible de aprendizaje por transferencia y aprendizaje por refuerzo permitiendo al modelo "M" adaptar su comportamiento cuando se conecta un nuevo módulo mientras mantiene el conocimiento previo adquirido. Finalmente, esto implica que se considera que el modelo "M" es capaz de realizar la tarea "T" bajo un número variable "N" de módulos.  The object of the invention conceives a possible combination of transfer learning and reinforcement learning allowing the "M" model to adapt its behavior when a new module is connected while maintaining the previous knowledge acquired. Finally, this implies that the "M" model is considered to be able to perform the "T" task under a variable number "N" of modules.
Una vez el modelo "M" ha realizado la tarea "T", los datos en relación a la salida generada se transfieren a, al menos, los actuadores del robot "R". Lo mismo es aplicable cuando se realizan múltiples tareas "T", T1 ", "T2", etc. Once the "M" model has completed the "T" task, the data in relation to the generated output are transferred to at least the actuators of the robot "R". The same is true when performing multiple tasks "T", T1 "," T2 ", etc.
El objeto de la invención permite el entrenamiento de robots modulares para conseguir una tarea dada bajo diferentes configuraciones de hardware mientras mantiene el mismo rendimiento. Para llevar a cabo dicha tarea, la invención define un modelo "M" compuesto de Redes Neuronales Artificiales, lógica programada o una combinación de ambas que hace uso de una descripción dinámica de los módulos "N" dentro del robot "R" (concretamente el modelo de robot "RM") para adaptar dinámicamente su comportamiento.  The object of the invention allows the training of modular robots to achieve a given task under different hardware configurations while maintaining the same performance. To carry out said task, the invention defines an "M" model composed of Artificial Neural Networks, programmed logic or a combination of both that makes use of a dynamic description of the "N" modules within the "R" robot (specifically the robot model "RM") to dynamically adapt its behavior.
DESCRIPCIÓN DE LOS DIBUJOS DESCRIPTION OF THE DRAWINGS
Para completar la descripción que se está realizando y para ayudar a una mejor comprensión de las características de la invención, de acuerdo con un ejemplo preferido de la realización práctica del mismo, se adjunta una serie de dibujos como parte integral de dicha descripción en los que, con un carácter ilustrativo y no limitativo, se ha representado lo siguiente:  To complete the description that is being made and to help a better understanding of the features of the invention, according to a preferred example of the practical realization thereof, a series of drawings are attached as an integral part of said description in which , with an illustrative and non-limiting nature, the following has been represented:
La figura 1.- Describe un diagrama de flujo que representa un modelo dado "M" de una realización preferida del método objeto de la invención.  Figure 1.- Describes a flow chart representing a given "M" model of a preferred embodiment of the method object of the invention.
La figura 2.- Muestra una representación en 3D de un robot modular que comprende tres articulaciones y cinco módulos, en los que no se han representado detalles de alimentación y comunicación.  Figure 2.- Shows a 3D representation of a modular robot comprising three joints and five modules, in which no power and communication details have been represented.
La figura 3.- Muestra una representación en 3D de un robot modular que comprende cinco módulos, en la que también se representa la tarea objetivo (T).  Figure 3.- Shows a 3D representation of a modular robot comprising five modules, in which the objective task (T) is also represented.
La figura 4.- Muestra una representación en 3D que un robot modular que comprende cinco módulos y tres articulaciones en el que se elimina la última articulación (dando como resultado cuatro módulos), en el que también se representa la tarea objetivo (T).  Figure 4.- Shows a 3D representation that a modular robot comprising five modules and three joints in which the last joint is eliminated (resulting in four modules), in which the objective task (T) is also represented.
La figura 5.- Muestra una representación en 3D del robot modular de la figura 4 después de que se haya eliminado la última articulación.  Figure 5.- Shows a 3D representation of the modular robot of figure 4 after the last joint has been removed.
REALIZACIÓN PREFERIDA DE LA INVENCIÓN PREFERRED EMBODIMENT OF THE INVENTION
En una realización preferida de la invención se proporciona un método para configurar un robot modular (R)—de aquí en adelante robot (R)— que comprende una serie de módulos (N) o módulos, definidos por actuadores (A1 , A2, A3). Cada módulo (N) podría incluir sus propios medios de cálculo y/o medios de detección adicionales siendo al menos proporcionados dichos módulos (N) con unidad de procesamiento de modo que puedan procesarse órdenes e instrucciones. La base del objeto de la invención se describe en la figura 1 en la que se compone un modelo (M) por Redes Neuronales Artificiales, lógica de programación o una combinación de ambas y se adapta para recibir un número de entradas (I) y un modelo de robot (RM) que es una representación de los módulos (N) que existen en un momento dado en el robot (R). El modelo (M) genera un cierto número de salidas (O) para uno o más actuadores del robot (R) de modo que el robot (R) lleve a cabo una tarea (T) dada independientemente del número de módulos (N) en el robot, si es físicamente factible. En una realización preferida de la invención el modelo de robot (RM) puede ser una lista que describe cada uno de los módulos (N) del robot (R) y cómo se relacionan entre sí; siendo dichos módulos (N) modelizados como módulos modelizados (Ν'). In a preferred embodiment of the invention there is provided a method for configuring a modular robot (R) - hereinafter robot (R) - comprising a series of modules (N) or modules, defined by actuators (A1, A2, A3 ). Each module (N) could include its own calculation means and / or additional detection means, at least said modules (N) being provided with a processing unit so that orders and instructions can be processed. The basis of the object of the invention is described in Figure 1 in which a model (M) is composed of Artificial Neural Networks, programming logic or a combination of both and is adapted to receive a number of inputs (I) and a robot model (RM) which is a representation of the modules (N) that exist at a given time in the robot (R). Model (M) generates a certain number of outputs (O) for one or more robot actuators (R) so that the robot (R) performs a given task (T) regardless of the number of modules (N) in the robot, if it is physically feasible. In a preferred embodiment of the invention the robot model (RM) may be a list describing each of the modules (N) of the robot (R) and how they relate to each other; said modules (N) being modeled as modeled modules (Ν ').
De acuerdo con las entradas (I) al menos uno de los módulos modelizados (Ν') del robot (R), concretamente los actuadores modelizados (Α1 ', Α2', A3') reaccionan a la entrada (I) citada. La entrada (I) puede comprender información relativa a la ejecución de una tarea (T) dada por el robot y/o información sensorial relativa a uno o más de un actuador (Α1 ', Α2', Α3'). La entrada (I) puede incluir información relativa a la ejecución de una tarea (T) dada, de modo que el robot (R) lleve a cabo la tarea (T).  According to the inputs (I) at least one of the modeled modules (Ν ') of the robot (R), specifically the modeled actuators (Α1', Α2 ', A3') react to the input (I) mentioned. The input (I) may comprise information related to the execution of a task (T) given by the robot and / or sensory information relating to one or more of an actuator (Α1 ', Α2', Α3 '). The input (I) can include information related to the execution of a given task (T), so that the robot (R) carries out the task (T).
Para llevar a cabo la tarea (T), los módulos (N) deben ejecutar en consecuencia una serie de órdenes y acciones; por ello en el mundo real, los actuadores (A1 , A2, A3) del robot (R) deberían ejecutar algunas órdenes y activar los movimientos respectivos de las mismas de modo que el robot (R) lleve a cabo la tarea (T).  To carry out the task (T), the modules (N) must execute a series of orders and actions accordingly; Therefore, in the real world, the actuators (A1, A2, A3) of the robot (R) should execute some orders and activate their respective movements so that the robot (R) performs the task (T).
Dada la tarea (T), puede realizarse por los actuadores (A1 , A2, A3) del robot (R) una pluralidad de movimientos posibles, los actuadores modelizados (A1 ', A2\ Given the task (T), a plurality of possible movements can be performed by the actuators (A1, A2, A3) of the robot (R), the modeled actuators (A1 ', A2 \
A3') correspondientes se usan para evaluar las combinaciones posibles de acciones requeridas para realizar la tarea (T). Corresponding A3 ') are used to evaluate the possible combinations of actions required to perform the task (T).
Por ello, el modelo (M) procesa las entradas (I) y genera las salidas (O) respectivas a partir de los actuadores modelizados (Α1 ', Α2', Α3')—no mostrados en las figuras— del modelo de robot (RM) de modo que lleva a cabo la tarea (T) dada independientemente del número de módulos (N) en el modelo de robot (RM), si es físicamente factible de acuerdo con o bien el modelo de robot (RM) o bien el robot (R) en sí. Obsérvese que el número de módulos (N) dentro del robot también se representa en el modelo de robot (RM).  Therefore, the model (M) processes the inputs (I) and generates the respective outputs (O) from the modeled actuators (Α1 ', Α2', Α3 ') - not shown in the figures - of the robot model ( RM) so that it performs the given task (T) regardless of the number of modules (N) in the robot model (RM), if it is physically feasible according to either the robot model (RM) or the robot (R) itself. Note that the number of modules (N) inside the robot is also represented in the robot model (RM).
El método de la invención puede implementarse de tal manera que, dado el robot (R) descrito en la figura 2 que comprende cuatro módulos (N), siendo por ello N=5 dichos módulos: un módulo de cerebro (B) que comprende una unidad de procesamiento central, el primer actuador (A1), el segundo actuador (A2), el tercer actuador (A3) y el efector terminal (EF). El robot (R) tiene un efector terminal (EF) en su extremo con un puntero láser hacia el espacio de trabajo. El objetivo, tarea (T) del robot (R) es que el efector terminal (EF) alcance un cierto punto en el espacio —marcado como "objetivo" en la figura 3— en el espacio de trabajo mediante la generación de movimientos en los actuadores (A1 , A2, A3), mientras se coordinan por el cerebro (B), que contiene el modelo (M) que comprende las instrucciones relacionadas con el comportamiento requerido para cada uno de los actuadores (A1 , A2, A3) de modo que el efector terminal (EF) alcance el cierto punto en el espacio—marcado como "objetivo" en la figura 3— . The method of the invention can be implemented in such a way that, given the robot (R) described in Figure 2 comprising four modules (N), whereby N = 5 said modules: a brain module (B) comprising a central processing unit, the first actuator (A1), the second actuator (A2), the third actuator (A3) and the terminal effector (EF). The robot (R) has a terminal effector (EF) at its end with a laser pointer towards the workspace. The objective, task (T) of the robot (R) is for the terminal effector (EF) to reach a certain point in the space - marked as "objective" in Figure 3 - in the workspace by generating movements in the actuators (A1, A2, A3), while being coordinated by the brain (B), which contains the model (M) comprising the instructions related to the behavior required for each of the actuators (A1, A2, A3) so that the terminal effector (EF) reaches the certain point in the space — marked as “objective” in Figure 3—.
En una realización preferida de la invención se elimina un módulo, por ejemplo y de acuerdo con la figura 4 se elimina el tercer actuador (A3); conduciendo a un robot (R) como el descrito en la figura 5, en el que el robot (R) resultante carece de la última articulación, concretamente el tercer actuador (A3) pero necesita llevar a cabo la tarea (T) dada, concretamente el alcance por el efector terminal (EF) del "objetivo" pero con cuatro módulos (N), de ahí N=4, en lugar de cinco módulos (N=5).  In a preferred embodiment of the invention a module is removed, for example and according to Figure 4 the third actuator (A3) is removed; leading to a robot (R) as described in figure 5, in which the resulting robot (R) lacks the last articulation, specifically the third actuator (A3) but needs to perform the given task (T), specifically the scope of the "target" by the terminal effector (EF) but with four modules (N), hence N = 4, instead of five modules (N = 5).
En una realización alternativa de la invención, el robot (R) que tiene inicialmente cuatro módulos (N=4) y dos actuadores (A1 , A2) se provee con un tercer actuador (A3), por lo que el robot (R) tiene ahora cinco módulos (de ahí N=5). Aún, el robot (R) necesita llevar a cabo la tarea (T) dada, concretamente el alcance por el efector terminal (EF) del "objetivo" pero con cinco módulos (N).  In an alternative embodiment of the invention, the robot (R) that initially has four modules (N = 4) and two actuators (A1, A2) is provided with a third actuator (A3), whereby the robot (R) has now five modules (hence N = 5). Still, the robot (R) needs to carry out the given task (T), specifically the scope by the terminal effector (EF) of the "target" but with five modules (N).
De ese modo, el método de la invención permite al robot (R) provisto con el tercer actuador (A3) adoptar el modelo (M) de modo que el robot (R) consiga el mismo "objetivo" y viceversa.  Thus, the method of the invention allows the robot (R) provided with the third actuator (A3) to adopt the model (M) so that the robot (R) achieves the same "target" and vice versa.
El método de la invención se basa en combinaciones factibles dadas de los módulos (N), en este caso: N=5 y N=4, dentro del robot (R) representado como el modelo de robot (RM).  The method of the invention is based on feasible combinations given of the modules (N), in this case: N = 5 and N = 4, within the robot (R) represented as the robot model (RM).
En general, dada la combinación físicamente factible de módulos (N) dentro del robot (R) representado como el modelo de robot (RM), el método de la invención comprende la realización de un proceso para la transformación del modelo de robot (RM) en una forma que permita representar características categóricas del modelo de robot (RM) en un formato que se comporte más eficientemente con algoritmos de clasificación y regresión típicamente usados en aprendizaje por refuerzo. El resultado se denomina Modelo de Robot Codificado (abreviadamente "ERM" del inglés "Encoded Robot Model"). A continuación el método proporcionado por la presente comprende el hallazgo de una codificación de tamaño fijo para el ERM (por ejemplo: en esta realización particular usando el método de Codificación de Tamaño-Fijo Olvido-Ordinal (Fixed-Size Ordinally- Forgetting Encoding)) denominado Modelo de Robot Codificado con Tamaño Fijo (abreviadamente "FERM") para entrenar o adaptar el modelo (M) a través de técnicas de aprendizaje por refuerzo usando una combinación de entradas (I) y de FERM. In general, given the physically feasible combination of modules (N) within the robot (R) represented as the robot model (RM), the method of the invention comprises performing a process for transforming the robot model (RM) in a way that allows to represent categorical characteristics of the robot model (RM) in a format that behaves more efficiently with classification and regression algorithms typically used in reinforcement learning. The result is called the Encoded Robot Model (abbreviated "ERM"). Next, the method provided herein comprises the finding of a fixed-size coding for the ERM (for example: in this particular embodiment using the Oblivion-Fixed-Fixed-Size Ordinary-Forgetting Encoding method) called Fixed Size Coded Robot Model (abbreviated "FERM") to train or adapt the model (M) through reinforcement learning techniques using a combination of inputs (I) and FERM.
El modelo (M) resultante sería capaz de ejecutar su tarea (T) (es decir alcanzar el "objetivo" para N=4 y N=5 tal como se ha descrito anteriormente. De modo similar, el modelo (M) podría adaptarse cuando se añaden (o eliminan) nuevos módulos (N) siempre que haya una factibilidad física para completar la tarea (T) con un robot (R) que comprenda dichos módulos (N), concretamente que alcance el "objetivo". Esto se consigue mediante la transferencia de la salida (O) a los módulos (N) del robot (R) como órdenes a ser ejecutadas para realizar la tarea (T).  The resulting model (M) would be able to execute its task (T) (that is, achieve the "objective" for N = 4 and N = 5 as described above. Similarly, the model (M) could be adapted when new modules (N) are added (or removed) provided there is a physical feasibility to complete the task (T) with a robot (R) that comprises said modules (N), specifically reaching the "objective." This is achieved by the transfer of the output (O) to the modules (N) of the robot (R) as orders to be executed to perform the task (T).

Claims

REIVINDICACIONES
1. Método para configurar un robot (R) modular, comprendiendo dicho robot una pluralidad de módulos (N) en el que cada uno de dichos módulos (N) comprende una unidad de procesamiento, estando caracterizado el método por que comprende: 1. Method for configuring a modular robot (R), said robot comprising a plurality of modules (N) in which each of said modules (N) comprises a processing unit, the method being characterized by comprising:
• generar un modelo de robot (RM) que es una representación de los módulos (N) del robot (R) en un momento dado,  • generate a robot model (RM) that is a representation of the modules (N) of the robot (R) at a given time,
• generar un modelo (M) que produce una salida (O), comprendiendo dicho modelo (M) a su vez al menos uno de entre: Redes Neuronales Artificiales, lógica programada, o una combinación de ambas.  • generate a model (M) that produces an output (O), said model (M) comprising at least one of: Artificial Neural Networks, programmed logic, or a combination of both.
• Alimentar el modelo (M) con:  • Feed the model (M) with:
• un número de entradas (I) que puede comprender información relativa a la ejecución de una tarea (T) dada por el robot, y • a number of inputs (I) that may comprise information related to the execution of a task (T) given by the robot, and
• el modelo de robot (RM), y • the robot model (RM), and
• generar una salida (O) por medio del modelo (M), comprendiendo dicha salida (O) órdenes relativas a la ejecución de la tarea (T) dada por los módulos (N) en el robot (R) en cualquier momento dado, y • generate an output (O) by means of the model (M), said output (O) comprising orders related to the execution of the task (T) given by the modules (N) in the robot (R) at any given time, Y
• transferir la salida (O) a los módulos (N) del robot (R) como órdenes a ser ejecutadas para realizar la tarea (T). • transfer the output (O) to the modules (N) of the robot (R) as orders to be executed to perform the task (T).
2. Método para configurar un robot (R) modular de acuerdo con la reivindicación 1 que comprende adicionalmente realizar un proceso de codificación para transformar las características categóricas del modelo de robot (RM) a un formato que se comporte más eficientemente con algoritmos de clasificación y regresión que generen un Modelo de Robot Codificado (ERM). 2. Method for configuring a modular robot (R) according to claim 1, further comprising performing a coding process to transform the categorical characteristics of the robot model (RM) to a format that behaves more efficiently with classification algorithms and regression that generate a Coded Robot Model (ERM).
3. Método para configurar un robot (R) modular de acuerdo con la reivindicación 2 en el que el proceso de codificación comprende hallar una codificación de tamaño fijo (FERM) para el Modelo de Robot Codificado (ERM) para entrenar o adaptar el modelo (M) a través de técnicas de aprendizaje por refuerzo usando una combinación de entradas (I) y codificación de tamaño fijo (FERM). 3. Method for configuring a modular robot (R) according to claim 2 wherein the coding process comprises finding a fixed size coding (FERM) for the Coded Robot Model (ERM) to train or adapt the model ( M) through reinforcement learning techniques using a combination of inputs (I) and fixed size coding (FERM).
4. Método para configurar un robot (R) modular de acuerdo con la reivindicación 3 en el que hallar la codificación de tamaño fijo para el ERM se lleva a cabo usando el método Fixed-Size Ordinally-Forgetting Encoding. 4. Method for configuring a modular robot (R) according to claim 3 wherein finding the fixed size coding for the ERM is carried performed using the Fixed-Size Ordinally-Forgetting Encoding method.
5. Método para configurar un robot (R) modular de acuerdo con la reivindicación 1 en el que los módulos (N) comprenden al menos uno de entre:5. Method for configuring a modular robot (R) according to claim 1 wherein the modules (N) comprise at least one of:
• sensores, para recibir y detectar el entorno, • sensors, to receive and detect the environment,
• actuadores, que permiten a los robots producir un cambio físico en el entorno,  • actuators, which allow robots to produce a physical change in the environment,
• módulos de comunicación, que proporcionan medios de interconexión o bien entre módulos o bien con sistemas externos, • communication modules, which provide means of interconnection either between modules or with external systems,
• módulos de procesamiento, para realizar las tareas más computacionalmente costosas dentro del robot, • processing modules, to perform the most computationally expensive tasks within the robot,
• módulos de alimentación que proporcionan energía global al robot modular, y  • power modules that provide global power to the modular robot, and
• módulos de interfaz de usuario (Ul), que proporcionan medios para interactuar con el robot modular.  • user interface modules (Ul), which provide means to interact with the modular robot.
PCT/ES2017/070781 2017-11-27 2017-11-27 Method for designing modular robots WO2018154153A2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/ES2017/070781 WO2018154153A2 (en) 2017-11-27 2017-11-27 Method for designing modular robots

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/ES2017/070781 WO2018154153A2 (en) 2017-11-27 2017-11-27 Method for designing modular robots

Publications (2)

Publication Number Publication Date
WO2018154153A2 true WO2018154153A2 (en) 2018-08-30
WO2018154153A3 WO2018154153A3 (en) 2018-11-01

Family

ID=61231282

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/ES2017/070781 WO2018154153A2 (en) 2017-11-27 2017-11-27 Method for designing modular robots

Country Status (1)

Country Link
WO (1) WO2018154153A2 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110297490A (en) * 2019-06-17 2019-10-01 西北工业大学 Heterogeneous module robot via Self-reconfiguration planing method based on nitrification enhancement

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE425112C (en) 1924-08-13 1926-02-11 Ernst Otto Baum Schraem rod with round or polygonal reversible cutting discs
US6995536B2 (en) 2003-04-07 2006-02-07 The Boeing Company Low cost robot manipulator
US20140019392A1 (en) 2012-06-01 2014-01-16 Brain Corporation Intelligent modular robotic apparatus and methods
US20150258683A1 (en) 2014-03-13 2015-09-17 Brain Corporation Trainable modular robotic apparatus and methods

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE4111354A1 (en) * 1991-04-09 1992-10-22 Bodenseewerk Geraetetech DEVICE FOR GUIDING THE END EFFECTOR OF A ROBOT ALONG A TARGET RAILWAY
US9384443B2 (en) * 2013-06-14 2016-07-05 Brain Corporation Robotic training apparatus and methods

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE425112C (en) 1924-08-13 1926-02-11 Ernst Otto Baum Schraem rod with round or polygonal reversible cutting discs
US6995536B2 (en) 2003-04-07 2006-02-07 The Boeing Company Low cost robot manipulator
US20140019392A1 (en) 2012-06-01 2014-01-16 Brain Corporation Intelligent modular robotic apparatus and methods
US20150258683A1 (en) 2014-03-13 2015-09-17 Brain Corporation Trainable modular robotic apparatus and methods

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110297490A (en) * 2019-06-17 2019-10-01 西北工业大学 Heterogeneous module robot via Self-reconfiguration planing method based on nitrification enhancement
CN110297490B (en) * 2019-06-17 2022-06-07 西北工业大学 Self-reconstruction planning method of heterogeneous modular robot based on reinforcement learning algorithm

Also Published As

Publication number Publication date
WO2018154153A3 (en) 2018-11-01

Similar Documents

Publication Publication Date Title
Ahmadzadeh et al. Modular robotic systems: Characteristics and applications
Beer The dynamics of brain–body–environment systems: A status report
Galluppi et al. Event-based neural computing on an autonomous mobile platform
Schmickl et al. ASSISI: mixing animals with robots in a hybrid society
Kuwabara et al. Timing-based control via echo state network for soft robotic arm
Iida et al. Biologically inspired robotics
Villacorta-Atienza et al. Compact internal representation of dynamic situations: neural network implementing the causality principle
WO2018154153A2 (en) Method for designing modular robots
Lele et al. Learning to walk: Bio-mimetic hexapod locomotion via reinforcement-based spiking central pattern generation
Tapia et al. Fast social-like learning of complex behaviors based on motor motifs
Dashevskiy et al. Survey of modular robots and developed embedded devices for constructive and computing components
Lomp et al. A software framework for cognition, embodiment, dynamics, and autonomy in robotics: cedar
Nguyen et al. Reaching development through visuo-proprioceptive-tactile integration on a humanoid robot-a deep learning approach
Roli et al. Attractor Landscape: A Bridge between Robotics and Synthetic Biology.
Rossi et al. Simultaneous versus incremental learning of multiple skills by modular robots
Asama et al. System principle on emergence of mobiligence and its engineering realization
Sofge et al. Challenges and opportunities of evolutionary robotics
WO2018172593A2 (en) Method for integrating new modules into modular robots, and robot component of same
Trenkwalder Computational resources of miniature robots: Classification and implications
Rubilar et al. Bio-inspired architecture for a reactive-deliberative robot controller
Doncieux et al. New horizons in evolutionary robotics: Extended contributions from the 2009 evoderob workshop
Kurtaj et al. The cerebellum: new computational model that reveals its primary function to calculate multibody dynamics conform to Lagrange-Euler formulation
Madhevan et al. Structures and characteristics in reconfigurable modular robots
Hauser Morphological computation–a potential solution for the control problem in soft robotics
Konarski et al. Mobile robot coordination using fear modeling algorithm

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17842358

Country of ref document: EP

Kind code of ref document: A2

122 Ep: pct application non-entry in european phase

Ref document number: 17842358

Country of ref document: EP

Kind code of ref document: A2