Generating robot behaviors in dynamic real-world situations generally requires the programming of multiple, often redundant degrees of freedom to meet multiple goals governing the desired motions. In this work, we propose a constraint-based controller specification methodology. A novel declarative language is used to combine semantically specialized building blocks into composite controllers. This description is automatically transformed at runtime into an executable form, which can automatically leverage multiple threads to parallelize computations whenever possible. Enabling runtime definition of controller topologies out of declarative descriptions not only reduces the work required to develop such controllers, but it also allows one to dynamically synthesize new controllers based on higher-level task planners or by user interaction through Graphical User Interfaces (GUIs). Our solution adds new functionality to the Robot Operating System (ROS)/ros_control ecosystem, where robot behaviors are typically achieved by deploying single-objective, off-the-shelf controllers for tasks like following joint trajectories, executing interpolated point-to-point motions in Cartesian space, or for basic compliant behaviors. Our proposed constraint-based framework enhances ros_control by providing the means to easily construct composite controllers from existing primary elements using our design language. Building on top of the ros_control infrastructure facilitates the usage of our controller with a wide range of supported robots and enables quick integration with the existing ROS ecosystem.

Easy, reliable and fast learning from demonstration approaches are desirable for task automation in construction sites. Unlike in classic assembly industry where the robot acts in a well-controlled setting, construction sites are unstructured and constantly changing environments. Therefore both workers and robots have to easily adapt to new working scenarios. In this paper we propose a method for automating the joint filling with mastic using a single human demonstration. The demonstration is performed by a human via teleoperation. The learning phase aims at estimating the appropriate tuning parameters of the admittance controller used to reproduce the human motion. Early stage laboratory testings presented in this document demonstrate the validity of the proposed learning scheme.

For the development of autonomous robotic systems, Dynamic Movement Primitives (DMP) and Artificial Potential Fields (APF) are two well known techniques. DMPs are a reference algorithm in robotics for one shot learning as they enable learning complex movements and generating smooth trajectories, while APF are outstanding in navigation and obstacle avoidance tasks. By integrating DMPs and APF, the task automation capability can be significantly enhanced, as the precision of DMPs combined with the reactive nature of APF promises, in theory, adaptability and efficiency for the learning algorithm. Despite the numerous papers discussing and reviewing both techniques independently, there is a lack of an objective comparison of the investigations combining both approaches. This paper aims to provide such a comprehensive literature analysis, using a homogenized mathematical formulation. Moreover, a categorization based on their application scope, the robots used and their characteristics is provided. Finally, open challenges in the combination of DMP and APF are discussed, highlighting further works that are worth conducting for improving the integration of both approaches.

Deep learning methods have revolutionized computer vision since the appearance of AlexNet in 2012. Nevertheless, 6 degrees of freedom pose estimation is still a difficult task to perform precisely. Therefore, we propose 2 ensemble techniques to refine poses from different deep learning 6DoF pose estimation models. The first technique, merge ensemble, combines the outputs of the base models geometrically. In the second, stacked generalization, a machine learning model is trained using the outputs of the base models and outputs the refined pose. The merge method improves the performance of the base models on LMO and YCB-V datasets and performs better on the pose estimation task than the stacking strategy.

In this paper, an innovative algorithm for averaging a set of multivariate time series with different lengths based on Constrained Dynamic Time Warping (CDTW) is proposed. This approach relies on the CDTW to provide the non-linear alignment of the multivariate time series, and employs the proposed Minimum Cost Averaging (MCA) technique to identify the optimum matches and get equal-length time series. MCA-CDTW is a task-agnostic approach that after selecting a reference curve, transforms the rest of the demonstrations in the set to obtain new curves that are time-aligned with the reference. From these transformed curves, not only the mean but also the signal variability can be directly extracted. This technique provides smooth mean curves even when there are large deviations between the demonstrations in the set, and still the complexity of the proposed algorithm is significantly reduced compared to other averaging techniques from the literature. When learning techniques are used to teach a motion to a robotic system, obtaining smooth trajectories is important to achieve good robotic behaviors. The new algorithm MCA-CDTW is tested and compared on two different databases: a literature database where humans move a robotic arm with kinaesthetic teaching, and a set of recordings of a teleoperated robotic arm performing laboratory manipulation. On both datasets, it is demonstrated that the new approach is providing smooth average trajectories.