Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8246949
A. V. Sureshbabu, M. Maggiali, G. Metta, A. Parmiggiani
This paper outlines the design of the hand of the R1 humanoid robot. The hand uses a completely plastic structure with embedded electronics. It has 2 actuated degrees of freedom (DOF) with 4 phalanges, coupling two phalanges to each degree of actuation. A novel series elastic module was developed within the hand. It is used in force sensing and protects the hand from impact loads. The series elastic module is designed, characterized and evaluated across the working range of the hand. The hand also has position sensors at all joints and tactile sensors for tactile feedback on its phalanges. The hand is completely self-contained with all control boards and motors housed within the structure. It is then tested and evaluated against user needs.
{"title":"Design of a force sensing hand for the R1 humanoid robot","authors":"A. V. Sureshbabu, M. Maggiali, G. Metta, A. Parmiggiani","doi":"10.1109/HUMANOIDS.2017.8246949","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8246949","url":null,"abstract":"This paper outlines the design of the hand of the R1 humanoid robot. The hand uses a completely plastic structure with embedded electronics. It has 2 actuated degrees of freedom (DOF) with 4 phalanges, coupling two phalanges to each degree of actuation. A novel series elastic module was developed within the hand. It is used in force sensing and protects the hand from impact loads. The series elastic module is designed, characterized and evaluated across the working range of the hand. The hand also has position sensors at all joints and tactile sensors for tactile feedback on its phalanges. The hand is completely self-contained with all control boards and motors housed within the structure. It is then tested and evaluated against user needs.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129226624","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8246932
Brandon J. DeHart, D. Kulić
There are two main goals for any mobile, bipedal system: locomotion and balance. These behaviors both require the biped to effectively move its center of mass (COM). In this work, we define an optimization framework which can be used to design a biped that maximizes its ability to move its COM, without having to define an associated controller or trajectory. We use angular momentum gain in our objective function, a measure of how efficiently a system can move its COM based on its physical properties. As a comparison, we also optimize the model using a cost of transport-based objective function over a set of trajectories and show that it provides similar results. However, the cost of transport calculation requires slow hybrid dynamics equations and hand-designed trajectories, whereas the angular momentum gain calculation requires only the joint space inertia matrix at each configuration of interest.
{"title":"Legged mechanism design with momentum gains","authors":"Brandon J. DeHart, D. Kulić","doi":"10.1109/HUMANOIDS.2017.8246932","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8246932","url":null,"abstract":"There are two main goals for any mobile, bipedal system: locomotion and balance. These behaviors both require the biped to effectively move its center of mass (COM). In this work, we define an optimization framework which can be used to design a biped that maximizes its ability to move its COM, without having to define an associated controller or trajectory. We use angular momentum gain in our objective function, a measure of how efficiently a system can move its COM based on its physical properties. As a comparison, we also optimize the model using a cost of transport-based objective function over a set of trajectories and show that it provides similar results. However, the cost of transport calculation requires slow hybrid dynamics equations and hand-designed trajectories, whereas the angular momentum gain calculation requires only the joint space inertia matrix at each configuration of interest.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123415126","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8246875
Daniel Tanneberg, Jan Peters, E. Rückert
Autonomous robots need to interact with unknown and unstructured environments. For continuous online adaptation in lifelong learning scenarios, they need sample-efficient mechanisms to adapt to changing environments, constraints, tasks and capabilities. In this paper, we introduce a framework for online motion planning and adaptation based on a bio-inspired stochastic recurrent neural network. By using the intrinsic motivation signal cognitive dissonance with a mental replay strategy, the robot can learn from few physical interactions and can therefore adapt to novel environments in seconds. We evaluate our online planning and adaptation framework on a KUKA LWR arm. The efficient online adaptation is shown by learning unknown workspace constraints sample-efficient within few seconds while following given via points.
{"title":"Efficient online adaptation with stochastic recurrent neural networks","authors":"Daniel Tanneberg, Jan Peters, E. Rückert","doi":"10.1109/HUMANOIDS.2017.8246875","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8246875","url":null,"abstract":"Autonomous robots need to interact with unknown and unstructured environments. For continuous online adaptation in lifelong learning scenarios, they need sample-efficient mechanisms to adapt to changing environments, constraints, tasks and capabilities. In this paper, we introduce a framework for online motion planning and adaptation based on a bio-inspired stochastic recurrent neural network. By using the intrinsic motivation signal cognitive dissonance with a mental replay strategy, the robot can learn from few physical interactions and can therefore adapt to novel environments in seconds. We evaluate our online planning and adaptation framework on a KUKA LWR arm. The efficient online adaptation is shown by learning unknown workspace constraints sample-efficient within few seconds while following given via points.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123078474","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8246934
J. Furukawa, Asuka Takai, J. Morimoto
In this study, we propose a databasedriven torque estimation approach for EMG-based robot control. For conventional EMG-based controllers, torque estimation models need to be carefully calibrated to control robots that have multiple degrees of freedom. However, such a calibration procedure requires significant effort and restricts the applications of EMG-based methods to practical situations. To cope with this issue, we use large-scale data acquired from other users to avoid the calibration process and propose collaborative filtering to estimate the joint torque of a new user by exploiting the previously derived relationships between the EMG signals and the joint torque of other users. To validate our proposed method, we compared the joint torque estimation performance with a standard linear conversion model. In our experiments, we controlled an upper-limb exoskeleton robot with the estimated joint torque where we used 16-ch electrodes to measure the EMG signals of subjects. In a comparison, our proposed method showed comparable control performance with the standard approach that requires a careful calibration process.
{"title":"Database-driven approach for Biosignal-based robot control with collaborative filtering","authors":"J. Furukawa, Asuka Takai, J. Morimoto","doi":"10.1109/HUMANOIDS.2017.8246934","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8246934","url":null,"abstract":"In this study, we propose a databasedriven torque estimation approach for EMG-based robot control. For conventional EMG-based controllers, torque estimation models need to be carefully calibrated to control robots that have multiple degrees of freedom. However, such a calibration procedure requires significant effort and restricts the applications of EMG-based methods to practical situations. To cope with this issue, we use large-scale data acquired from other users to avoid the calibration process and propose collaborative filtering to estimate the joint torque of a new user by exploiting the previously derived relationships between the EMG signals and the joint torque of other users. To validate our proposed method, we compared the joint torque estimation performance with a standard linear conversion model. In our experiments, we controlled an upper-limb exoskeleton robot with the estimated joint torque where we used 16-ch electrodes to measure the EMG signals of subjects. In a comparison, our proposed method showed comparable control performance with the standard approach that requires a careful calibration process.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"72 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126063633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8246972
Robin Rasch, S. Wachsmuth, Matthias König
To enable personal robots to operate in human spaces, it is necessary that robots support everyday tasks like handing over an object. Studies show that robots have to move and behave human-like, to improve social acceptance. Therefore, it is necessary to study and model human movements. This paper studies and analyses the movements of arms during hand-over between two persons in order to extract the characteristic features (elementary movements of joints, duration, angular and linear velocities, etc.). In the present study, we are using inertial measurement units with 6-axis (gyroscope and accelerometer) on wrist, elbow and shoulder to measure the movements and evaluate them. Our results show a general movement pattern for hand-overs between humans with two variants of twisting the elbow. The results of our study provide a basis for developing a human-like handover controller for humanoid robot systems or human like manipulators.
{"title":"Understanding movements of hand-over between two persons to improve humanoid robot systems","authors":"Robin Rasch, S. Wachsmuth, Matthias König","doi":"10.1109/HUMANOIDS.2017.8246972","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8246972","url":null,"abstract":"To enable personal robots to operate in human spaces, it is necessary that robots support everyday tasks like handing over an object. Studies show that robots have to move and behave human-like, to improve social acceptance. Therefore, it is necessary to study and model human movements. This paper studies and analyses the movements of arms during hand-over between two persons in order to extract the characteristic features (elementary movements of joints, duration, angular and linear velocities, etc.). In the present study, we are using inertial measurement units with 6-axis (gyroscope and accelerometer) on wrist, elbow and shoulder to measure the movements and evaluate them. Our results show a general movement pattern for hand-overs between humans with two variants of twisting the elbow. The results of our study provide a basis for developing a human-like handover controller for humanoid robot systems or human like manipulators.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"44 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126991603","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8239534
Tobia Marcucci, Robin Deits, M. Gabiccini, A. Bicchi, Russ Tedrake
Feedback control of robotic systems interacting with the environment through contacts is a central topic in legged robotics. One of the main challenges posed by this problem is the choice of a model sufficiently complex to capture the discontinuous nature of the dynamics but simple enough to allow online computations. Linear models have proved to be the most effective and reliable choice for smooth systems; we believe that piecewise affine (PWA) models represent their natural extension when contact phenomena occur. Discrete-time PWA systems have been deeply analyzed in the field of hybrid Model Predictive Control (MPC), but the straightforward application of MPC techniques to complex systems, such as a humanoid robot, leads to mixed-integer optimization problems which are not solvable at real-time rates. Explicit MPC methods can construct the entire control policy offline, but the resulting policy becomes too complex to compute for systems at the scale of a humanoid robot. In this paper we propose a novel algorithm which splits the computational burden between an offline sampling phase and a limited number of online convex optimizations, enabling the application of hybrid predictive controllers to higher-dimensional systems. In doing so we are willing to partially sacrifice feedback optimality, but we set stability of the system as an inviolable requirement. Simulation results of a simple planar humanoid that balances by making contact with its environment are presented to validate the proposed controller.
{"title":"Approximate hybrid model predictive control for multi-contact push recovery in complex environments","authors":"Tobia Marcucci, Robin Deits, M. Gabiccini, A. Bicchi, Russ Tedrake","doi":"10.1109/HUMANOIDS.2017.8239534","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8239534","url":null,"abstract":"Feedback control of robotic systems interacting with the environment through contacts is a central topic in legged robotics. One of the main challenges posed by this problem is the choice of a model sufficiently complex to capture the discontinuous nature of the dynamics but simple enough to allow online computations. Linear models have proved to be the most effective and reliable choice for smooth systems; we believe that piecewise affine (PWA) models represent their natural extension when contact phenomena occur. Discrete-time PWA systems have been deeply analyzed in the field of hybrid Model Predictive Control (MPC), but the straightforward application of MPC techniques to complex systems, such as a humanoid robot, leads to mixed-integer optimization problems which are not solvable at real-time rates. Explicit MPC methods can construct the entire control policy offline, but the resulting policy becomes too complex to compute for systems at the scale of a humanoid robot. In this paper we propose a novel algorithm which splits the computational burden between an offline sampling phase and a limited number of online convex optimizations, enabling the application of hybrid predictive controllers to higher-dimensional systems. In doing so we are willing to partially sacrifice feedback optimality, but we set stability of the system as an inviolable requirement. Simulation results of a simple planar humanoid that balances by making contact with its environment are presented to validate the proposed controller.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130345051","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8246882
Shunichi Nozawa, Masaki Murooka, Shintaro Noda, Kunio Kojima, Yuta Kojio, Youhei Kakiuchi, K. Okada, M. Inaba
In the case of object manipulation, a humanoid robot should consider the two-body problem between the object and the robot. To achieve this, the motion planner and the controller must satisfy constraints among the robot, the object, and environments. In addition, the objecfs properties such as mass properties and friction are not known a priori and the robot must obtain this information based on sensor feedback. In this paper, we propose a method for uniform humanoid manipulation of an unknown object by estimating objectenvironment constraints based on changes in the robofs force sensor measurements. The proposed method supports various types of manipulation (lifting, pushing, pivoting), various robot contacts (single-armed, dual-armed, full-body), multi-robot cooperative manipulation, and motion on a movable object. We evaluate the proposed method through experiments involving manipulation of large and heavy objects using life-sized real robots.
{"title":"Unified humanoid manipulation of an object of unknown mass properties and friction based on online constraint estimation","authors":"Shunichi Nozawa, Masaki Murooka, Shintaro Noda, Kunio Kojima, Yuta Kojio, Youhei Kakiuchi, K. Okada, M. Inaba","doi":"10.1109/HUMANOIDS.2017.8246882","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8246882","url":null,"abstract":"In the case of object manipulation, a humanoid robot should consider the two-body problem between the object and the robot. To achieve this, the motion planner and the controller must satisfy constraints among the robot, the object, and environments. In addition, the objecfs properties such as mass properties and friction are not known a priori and the robot must obtain this information based on sensor feedback. In this paper, we propose a method for uniform humanoid manipulation of an unknown object by estimating objectenvironment constraints based on changes in the robofs force sensor measurements. The proposed method supports various types of manipulation (lifting, pushing, pivoting), various robot contacts (single-armed, dual-armed, full-body), multi-robot cooperative manipulation, and motion on a movable object. We evaluate the proposed method through experiments involving manipulation of large and heavy objects using life-sized real robots.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"136 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114213151","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8246887
Kunihiro Ogata, Hideyuki Tanaka, Y. Matsumoto
Elderly individuals are likely to develop locomotive disorders such as osteoarthritis or osteoporosis. This increases the risk of falls and makes independent movement difficult. Elderly individuals should better understand walking function to extend their healthy life. We therefore propose a new method for estimating hand and foot reaction forces using only visual markers and a monocular camera. When humans contact the environment with their hands, their hand and feet positions define a convex hull. A proposed “ generalized zero moment point ” is projected on this convex hull, which is approximated as a line or plane, and the distance between this point and each contact point is calculated. Reaction forces are calculated based on the ratios of these distances. Evaluation experiments show high agreement between estimated and measured forces of both hands and feet, confirming the validity of the proposed algorithm.
{"title":"Estimating hand and foot reaction forces based on a generalized zero moment point for rehabilitation assist system","authors":"Kunihiro Ogata, Hideyuki Tanaka, Y. Matsumoto","doi":"10.1109/HUMANOIDS.2017.8246887","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8246887","url":null,"abstract":"Elderly individuals are likely to develop locomotive disorders such as osteoarthritis or osteoporosis. This increases the risk of falls and makes independent movement difficult. Elderly individuals should better understand walking function to extend their healthy life. We therefore propose a new method for estimating hand and foot reaction forces using only visual markers and a monocular camera. When humans contact the environment with their hands, their hand and feet positions define a convex hull. A proposed “ generalized zero moment point ” is projected on this convex hull, which is approximated as a line or plane, and the distance between this point and each contact point is calculated. Reaction forces are calculated based on the ratios of these distances. Evaluation experiments show high agreement between estimated and measured forces of both hands and feet, confirming the validity of the proposed algorithm.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124934250","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8246954
E. Hoffman, A. Rocchi, Arturo Laurenzi, N. Tsagarakis
In this paper we present OpenSoT, an open-source, recently developed software library, that can be used to solve robotics related control problems in a flexible and easy way. OpenSoT includes high-level interfaces to state-of-the-art algorithms for kinematic/dynamic modelling, quadratic programming optimization, cost functions and constraints specification. OpenSoT is implemented in C++ and permits rapid prototyping of controllers for fixed or floating base, highly redundant robots such as (but not limited to) manipulators and humanoids. We discuss the use of OpenSoT from the perspective of the developer and the user, leaving out details on the implementation of the tool. We demonstrate how the software can be used with two examples: control of a redundant humanoid robot through simple inverse kinematics schemes and contact forces optimization.
{"title":"Robot control for dummies: Insights and examples using OpenSoT","authors":"E. Hoffman, A. Rocchi, Arturo Laurenzi, N. Tsagarakis","doi":"10.1109/HUMANOIDS.2017.8246954","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8246954","url":null,"abstract":"In this paper we present OpenSoT, an open-source, recently developed software library, that can be used to solve robotics related control problems in a flexible and easy way. OpenSoT includes high-level interfaces to state-of-the-art algorithms for kinematic/dynamic modelling, quadratic programming optimization, cost functions and constraints specification. OpenSoT is implemented in C++ and permits rapid prototyping of controllers for fixed or floating base, highly redundant robots such as (but not limited to) manipulators and humanoids. We discuss the use of OpenSoT from the perspective of the developer and the user, leaving out details on the implementation of the tool. We demonstrate how the software can be used with two examples: control of a redundant humanoid robot through simple inverse kinematics schemes and contact forces optimization.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"101 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122694924","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2017-11-01DOI: 10.1109/HUMANOIDS.2017.8246910
F. Ledezma, S. Haddadin
Modeling physical systems with neural networks (NN) requires expert architects to determine the best number of nodes, layers and activation functions. For complex systems, such as articulated robots, reported results are limited in accuracy and generalization capabilities. In this work, we introduce the concept FOPnet. It is based on first-order principles and system knowledge to determine topologies of parametrized operator networks that accurately model input-output mappings of physical systems. These topologies consist of meaningful building elements and connections as well as a reduced number of parameters that describe the variables' interdependencies. In this way, learning speed is boosted and the model's accuracy, precision and generalization power improved. We apply the methodology to a 7 degrees-of-freedom LWR4 manipulator and discuss the estimation and generalization capabilities of the network. Results are compared to conventional Feed Forward NN as well as a state-of-the-art Deep Recurrent NN. For the considered complex robot dynamics FOPnet was able to achieve a seven orders of magnitude smaller generalization RMSE.
{"title":"First-order-principles-based constructive network topologies: An application to robot inverse dynamics","authors":"F. Ledezma, S. Haddadin","doi":"10.1109/HUMANOIDS.2017.8246910","DOIUrl":"https://doi.org/10.1109/HUMANOIDS.2017.8246910","url":null,"abstract":"Modeling physical systems with neural networks (NN) requires expert architects to determine the best number of nodes, layers and activation functions. For complex systems, such as articulated robots, reported results are limited in accuracy and generalization capabilities. In this work, we introduce the concept FOPnet. It is based on first-order principles and system knowledge to determine topologies of parametrized operator networks that accurately model input-output mappings of physical systems. These topologies consist of meaningful building elements and connections as well as a reduced number of parameters that describe the variables' interdependencies. In this way, learning speed is boosted and the model's accuracy, precision and generalization power improved. We apply the methodology to a 7 degrees-of-freedom LWR4 manipulator and discuss the estimation and generalization capabilities of the network. Results are compared to conventional Feed Forward NN as well as a state-of-the-art Deep Recurrent NN. For the considered complex robot dynamics FOPnet was able to achieve a seven orders of magnitude smaller generalization RMSE.","PeriodicalId":143992,"journal":{"name":"2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids)","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123762977","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：