Fig 1 - uploaded by Jens Lundell
Content may be subject to copyright.
Source publication
Policy search reinforcement learning allows robots to acquire skills by themselves. However, the learning procedure is inherently unsafe as the robot has no a-priori way to predict the consequences of the exploratory actions it takes. Therefore, exploration can lead to collisions with the potential to harm the robot and/or the environment. In this...
Contexts in source publication
Context 1
... for ensuring safe exploration in policy search (Section II) consisting of a task-prioritized inverse kinematics framework (Section II-A) which facilitates safe learning of a time-invariant policy (Section II-B); • an experimental evaluation (Section III) for a simulated reaching skill (Section III-A) and grasping skill on the platform shown in Fig. 1 (Section III-B) demonstrating that our method increases both safety, by removing the collision risk, and learning rate, by reducing the number of roll-outs before ...
Context 2
... validate our approach by means of two illustrative examples: reaching (Section III-A) and grasping (Section III-B). To this end, we use one of the 7 DOF arms of the platform depicted in Fig. 1. The reaching experiment was simulated in Gazebo, while the grasping experiment was executed on the actual ...
Context 3
... this experiment, the goal is to demonstrate our method working on a real robot in the sense that it learns policies for grasping nontrivial objects after few roll-outs. To achieve this, the experiment is to grasp the toy and box displayed in Fig. 4 with the robot shown in Fig. 1. The results presented in the previous section indicate that prior data in form of additional tasks allows for both faster and safer policy search. Therefore, we attempt to maximize the amount of prior knowledge in order to reduce the search space. To this end, we use tasks forming a so-called grasping envelope as defined in our ...
Similar publications
A multi-link cable-driven robot (MCDR) usually has a large number of redundant actuating cables due to its modular cable routing scheme. To reduce the number of actuating cables while keeping the advantages of the modular MCDRs, a hybrid modular cable routing method is proposed, in which some actuating cables are co-shared by adjacent cable-driven...
Citations
... The motion planning and control of the robot are completed in MoveIt, an open-source project in ROS. In this case, we modified an open-source unified robot description format (URDF) model of YuMi provided by R. Krug et al. [35]. Because YuMi's manipulators and grippers are controlled independently based on different IP addresses, we divided the whole robot into four motion planning groups after URDF remodeling: left arm, right arm, left hand and right hand. ...
As a senile chronic, progressive and currently incurable disease, dementia has an enormous impact on society and life quality of the elderly. The development of teleoperation technology has changed the traditional way of care delivery and brought a variety of novel applications for dementia care. In this paper, a telerobotic system is presented which gives the caregivers the capability of assisting dementia elderly remotely. The proposed system is composed of a dual-arm collaborative robot (YuMi) and a wearable motion capture device. The communication architecture is achieved by the robot operation system (ROS). The position-orientation data of the operator’s hand are obtained and used to control the YuMi robot. Besides, a path-constrained mapping method is designed for motion trajectory tracking between the robot and the operator in the progress of teleoperation. Meanwhile, corresponding experiments are conducted to verify the performance of the trajectory tracking using the path-constrained mapping method. Results show that the position tracking deviation between the trajectory of the operator and the robot measured by dynamic time warping distance is 1.05 mm at the sampling frequency of 7.5 Hz. Moreover, the practicability of the proposed system was verified by teleoperating the YuMi robot to pick up a medicine bottle and further demonstrated by assisting an elderly woman in picking up a cup remotely. The proposed telerobotic system has potential utility for improving the life quality of dementia elderly and the care effect of their caregivers.
