In many physical human-robot interaction scenarios, for successful completion of the tasks, robots should he able to recognize the human partner's intention. One of such scenarios that is studied in this letter is...
详细信息
In many physical human-robot interaction scenarios, for successful completion of the tasks, robots should he able to recognize the human partner's intention. One of such scenarios that is studied in this letter is the collaborative task of carrying an object by a human-humanoid pair in which the humanoid should he able to interpret specific human partner's intentions (e.g., start/stop-walking, accelerate, etc.) only through haptic feedback. To address this problem, we first performed human-human experiments and obtained a multiclass classifier (with more than 90% of accuracy) for human intention detection using, as features, arm position relative to the shoulder and interaction forces. The results of the multiclass classification, without any modifications, have been used to develop an interlimb coordinator that was integrated in a modular control architecture into human-robot experiments. The interlimb coordinator receives the sensory data of the upper-body and sends appropriate commands (including start/stop-walking, accelerate, and decelerate commands) to the lower body controller, which is responsible for achieving a stable walking gait. This modular control approach is successfully tested in the human-humanoid experiments with the COMAN robot.
In this letter, we present a general approach to automatically visual servo control the position and shape of a deformable object whose deformation parameters are unknown. The servo control is achieved by online learn...
详细信息
In this letter, we present a general approach to automatically visual servo control the position and shape of a deformable object whose deformation parameters are unknown. The servo control is achieved by online learning a model mapping between the robotic end-effector's movement and the object's deformation measurement. The model is learned using the Gaussian process regression (GPR) to deal with its highly nonlinear property, and once learned, the model is used for predicting the required control at each time step. To overcome GPR's high computational cost while dealing with long manipulation sequences, we implement a fast online GPR by selectively removing uninformative observation data from the regression process. We validate the performance of our controller on a set of deformable object manipulation tasks and demonstrate that our method can achieve effective and accurate servo control for general deformable objects with awide variety of goal settings.
Reinforcement learning (RL) is a suitable approach for controlling systems with unknown or time-varying dynamics. RL in principle does not require a model of the system, but before it learns an acceptable policy, it n...
详细信息
ISBN:
(纸本)9781538630815
Reinforcement learning (RL) is a suitable approach for controlling systems with unknown or time-varying dynamics. RL in principle does not require a model of the system, but before it learns an acceptable policy, it needs many unsuccessful trials, which real robots usually cannot withstand. It is well known that RL can be sped up and made safer by using models learned online. In this paper, we propose to use symbolic regression to construct compact, parsimonious models described by analytic equations, which are suitable for realtime robot control. Single node genetic programming (SNGP) is employed as a tool to automatically search for equations fitting the available data. We demonstrate the approach on two benchmark examples: a simulated mobile robot and the pendulum swing-up problem;the latter both in simulations and real-time experiments. The results show that through this approach we can find accurate models even for small batches of training data. Based on the symbolic model found, RL can control the system well.
In contact-based manipulations, the effects of the robot's actions change as contacts are made or broken. For example, if a robot applies an increasing upward force to an object, then the force will eventually ove...
详细信息
In contact-based manipulations, the effects of the robot's actions change as contacts are made or broken. For example, if a robot applies an increasing upward force to an object, then the force will eventually overcome the object's weight and break the object-table contact. The robot can subsequently raise or lower the height of the object. The transition from resting on the table to not being in contact with the table is an example of a mode switch. The conditions for this mode switch depend on the mass of the object being manipulated. By modeling the mode switch, the robot can estimate the mass of the object based on the conditions when the mode switch occurs. The robot can also use the model to predict when the object will break contact given its mass. We evaluated four different autoregressive hidden Markov models for representing manipulations with mass-dependent mode switches. The models were successfully evaluated on pushing and lifting tasks. The evaluations show that the predicted object trajectories and estimated object masses are more accurate when using models that interpolate between different masses, and that consider the observed state for estimating the mode switches.
Reinforcement learning (RL) is a suitable approach for controlling systems with unknown or time-varying dynamics. RL in principle does not require a model of the system, but before it learns an acceptable policy, it n...
详细信息
Reinforcement learning (RL) is a suitable approach for controlling systems with unknown or time-varying dynamics. RL in principle does not require a model of the system, but before it learns an acceptable policy, it needs many unsuccessful trials, which real robots usually cannot withstand. It is well known that RL can be sped up and made safer by using models learned online. In this paper, we propose to use symbolic regression to construct compact, parsimonious models described by analytic equations, which are suitable for real-time robot control. Single node genetic programming (SNGP) is employed as a tool to automatically search for equations fitting the available data. We demonstrate the approach on two benchmark examples: a simulated mobile robot and the pendulum swing-up problem;the latter both in simulations and real-time experiments. The results show that through this approach we can find accurate models even for small batches of training data. Based on the symbolic model found, RL can control the system well.
暂无评论