Research On Modular Robots For Quick Testing Of Evolved Robot Morphologies

The modular robot is a robot system which is composed of several modules with certain structures and functions,each module is an independent motion unit,and is able to to connect and communicate with neighboring modules.Unlike the traditional robot with fixed configuration,the most obvious advantages of modular robots are their strong deformability and environmental adaptability.According to different tasks and environment,modules can be assembled in various configurations leading to different robot morphologies.In addition,modular robots offer a benefit to robotics,especially evolutionary robotics,since they can be easily reconfigured to form a distinct morphology.Moreover,identical modules are easy to produce.Considering these advantages,modular robots have been applied to many fields,ranging from education and commercial tools to serarch and rescue.The basic component of modular robots is the module,in which the connection mechanism,power and data transmission mode,as well as the movement capability are the key factors of modular robot's motion control.However,on the current existing modular robotic platforms,mechanical design,connection mechanism design and assembly process of the module are relatively complicated.The difficulty in programming,debugging and control of the module is high,and moreover,the openness of the module design is unfriendly,resulting in the time and consumable cost of the transplantation process: when an evolutionary algorithm is used to design the morphological and control parameters of modular robots,most of evolved morphologies of robots can only be evaluated in the simulation,and cannot be quickly tested on the physical modular robots based on the evolved morphologies obtained in the simulation.This problem leads to the low efficiency of evolution experiment,reduces the design efficiency of acquiring high-performance modular robots,and increases the difficulty of debugging of modular robots.Moreover,it is not conducive to the promotion of modular robotics technology.To address this issue,firstly the design criteria of modules for quick tesing evolved morphologies of modular robotics are analyzed,and then a new type of open and easy-to-configure modular robot system EMERGE is proposed based on this criteria.On the basis of one degree of freedom in one module,a cuboid shape design using four connection faces is designed after the comparison and analysis of different configuration schemes of the number of connection faces of the EMERGE module,the number of configurations and external structures.Three of connection faces are connected to the motor's shaft and the other one attached to motor's chassis in order to form a multi-structure stable filling structure.A magnetic connection mechanism has been designed,symmetrically distributed circular magnets are mounted on each connection face,which has the ability to connect and disconnect quickly and reliably from the adjacent modules without energy consumption.Moreover,spring-compressed pins are used to effectively realize shareing of the power and communications through the connector.Most components of the module are commercially standardized devices,which are easy to be assembled.In addition,the mechanical design of other componets is open source,and are 3D printed,resulting in easy modifying and avoiding customizing at the factory,reducing the costs effectively.Each connection face of the module is equipped with a proximity sensor,which has the ability to sense the surrounding environment and expands the types of tasks performed by the modular robot.The standard and control-friendly Dynamixel motors(AX-12A)and their accessories are employed.The control system of the module is designed in distributed and centralized ways,which can be flexible switched according to different control requirements.In addition,the hardware circuit and software code of the control system are designed to be completely open source,which are easy to use and modify for everyone.This module is conducive to improve the design efficiency of high-performance modular robots and reduce the difficulty of actual debugging.In addition,it lays the foundation for the experimental study of robotic arm assembling of different morphologies of modular robots.The number and relative position of available connection faces in a module are important constituent elements of the module morphology,and it is the basis of the connection between adjacent modules and is also a direct manifestation of the diversity of robot morphologies.The modular robot is a complex robot system composed of multiple modules,so generally evolutionary algorithms are used to search for the optimal performance of a modular robot for a given task from the search space.As the number of connection faces in a module changes,the size of the search space also changes in the same direction.Therefore,this is related to the ability to search for the good performance of modular robots from a large number of different configurations of robots.However,with the increase of the number of connection faces,the size of the search space will gradually increase,the optimization process will be more time consuming,and it may also fall into a local optimal solution.Conversely,reducing the search space can improve the search quality of evolutionary algorithms.If the other evolution parameters are the same,only the number of connection faces in a module is considered,then the size of the search space is fixed.Most researches focus on searching for the good performance of modular robots under the fixed number of connection faces in a module,however,no consideration has been given to the effect of the change in the number of available connecting faces in a module on the efficiency of obtaining a high-performance of modular robots.In addition,whether the global optimal solution can be found depends on the convergence of the algorithm.The convergence of the algorithm is based on infinite iteration time.However,the computational iteration time in practical applications is limited and it is difficult to guide algorithm design and improvement in practice.It can be seen that how to reduce the size of the search space in the case of approximate convergence of the algorithm to find a good performance of modular robot in a shorter computing time is more practical.To this end,we takes the EMERGE modular robot as the research object and proposes an approach based on evolutionary design to evaluate module morphology of the modular robot.Firstly,the approach defines the number of connection faces and their relative positions as the morphological parameters,and the classification of different module morphologies is analyzed.Afterwards,an improved evolutionary strategy with overlapping detection function is proposed,which considers the number of configurations and modules and their feasibility simultanelously,to effectively eliminate the modular robots with overlapped collisions.The morphology and control of robots are evolved for a locomotion task in a robotic simulation platform to obtain modular robots composed of different types of EMERGE module.Then,statistical analysis of the simulation results is provided,which shows the best module morphology and the robot assembled by the modules.Finally,the simulated robots are transferred to the physical robots for testing,which verify the effectiveness of the approach.The results show that by using a specific type of module morphology drastically improve the efficiency of acquiring good permormance of robotic morphologies.In addition,when designing a modular robot for a given task,limiting the number of connection faces in a module also helps to reduce the size of the search space,thereby to obtain a high-performance modular robot in a shorter calculation time.When evolutionary mechanism are used to estimate the performance of modular robots in a robotic simulation platform,the simulated performance maybe good,but when it is transplanted to the physical robot,the performance is degraded,that is,the reality gap occurs.How to reduce the difference of robot performance between simulation and physical is of great significance to the practical application of modular robots.Some studies have achieved preliminary results,but the prerequisites for using are too harsh,this is reflected in several aspects: the scope of application is limited,and the performance difference is still large.Therefore,there is an urgent need to study a transplantation approach which is easy to use and has a significant improvement in performance.To this end,a transplantation approach is proposed that takes the fitness function value and transferability as double evolutionary goals,and it also improves the transferability of the robot and performance of the robot simultanelously.Firstly,the trade-off relationship between fitness function value and transferability is analyzed,and the characteristic meaning of the optimal solution is given.Afterwards,the characteristics of different spatial interpolation methods are analyzed and compared.The characteristics and deficiencies of the traditional inverse distance weighting method are analyzed in detail.Therefore,an improved inverse distance weighting method based on the relationship of sample orientation is proposed to make the sample points evenly distributed and also adapt to configure the power exponent simultanelously,which improves the efficiency and accuracy of the interpolation.Then this apporoach used to establish the surrogate model of difference between simulated and physical behaviour of the modular robot.Based on the surrogate model technology,the algorithm of the transplantation apporoach is studied.Finally,the transferability of modular robots assembled by different morphology types of EMERGE module for locomotion task is studied,and compared with the transplantation apporoach based on traditional inverse distance weight method.The results show that this approach is easy to use,and can effectively reduce the differences in fitness function values and action behaviors between simulated and physical robots.In order to verify the reliability of the EMERGE modular robot system,a physical experiment system has been established.Experiments of the load capacity of the connection mechanism,the configuration assemble,and the motion planning of the ring configuration,the chain configuration,the quadruped configuration,and the proximity sensing capability have benn performed.The results have proved that the EMERGE system has strong capabilities of a reasonable structure,good flexibility,reliability and stability during the movement process,has the ability to quickly test different evolved morphologies of modular robots,and also verifies the effectiveness of the motion planning method.