Rosy: An elegant language to teach the pure reactive nature of robot programming

Robotics is very appealing and is long recognized as a great way to teach programming, while drawing inspiring connections to other branches of engineering and science such as maths, physics or electronics. Although this symbiotic relationship between robotics and programming is perceived as largely beneficial, educational approaches often feel the need to hide the underlying complexity of the robotic system, but as a result fail to transmit the reactive essence of robot programming to the roboticists and programmers of the future. This paper presents Rosy, a novel language for teaching novice programmers through robotics. Its functional style is both familiar with a high-school algebra background and a materialization of the inherent reactive nature of robotic programming. Working at a higher-level of abstraction also teaches valuable design principles of decomposition of robotics software into collections of interacting controllers. Despite its simplicity, Rosy is completely valid Haskell code compatible with the ROS~ecosystem. We make a convincing case for our language by demonstrating how non-trivial applications can be expressed with ease and clarity, exposing its sound functional programming foundations, and developing a web-enabled robot programming environment.

DQN as an alternative to Market-based approaches for Multi-Robot processing Task Allocation (MRpTA)

Multi-robot task allocation (MRTA) problems require that robots make complex choices based on their understanding of a dynamic and uncertain environment. As a distributed computing system, the Multi-Robot System (MRS) must handle and distribute processing tasks (MRpTA). Each robot must contribute to the overall efficiency of the system based solely on a limited knowledge of its environment. Market-based methods are a natural candidate to deal processing tasks over a MRS but recent and numerous developments in reinforcement learning and especially Deep Q-Networks (DQN) provide new opportunities to solve the problem. In this paper we propose a new DQN-based method so that robots can learn directly from experience, and compare it with Market-based approaches as well with centralized and purely local solutions. Our study shows the relevancy of learning-based methods and also highlight research challenges to solve the processing load-balancing problem in MRS.

Packing Planning and Execution Considering Arrangement Rules

To send a consumer the ordered products in e-commerce, it is necessary to pack the products into a container. When packing, we need to consider the arrangement of the products to avoid damaging them during transportation by using the know-how of packing. In this paper, we propose a packing robot system that plans the arrangement and then executes it. To develop this robot system, we employ two ideas. The first idea is to split the whole process into the arrangement of the objects and robot motion generation. The arrangement is regarded as a 3D bin-packing problem. The second idea is to use the heuristics of the packing so that objects are stacked from the bottom. Since the space above the target position is free from collisions, the strategy to approach from the top reduces the possibility of a collision. In the experiments, we succeeded in arranging eleven objects satisfying the assigned rules and packing ten of the eleven objects using the robot.

Reactive Navigation Framework for Mobile Robots by Heuristically Evaluated Pre-sampled Trajectories

This paper describes and analyzes a reactive navigation framework for mobile robots in unknown environments. The approach does not rely on a global map and only considers the local occupancy in its robot-centered 3D grid structure. The proposed algorithm enables fast navigation by heuristic evaluations of pre-sampled trajectories on-the-fly. At each cycle, these paths are evaluated by a weighted cost function, based on heuristic features such as closeness to the goal, previously selected trajectories, and nearby obstacles. This paper introduces a systematic method to calculate a feasible pose on the selected trajectory, before sending it to the controller for the motion execution. Defining the structures in the framework and providing the implementation details, the paper also explains how to adjust its offline and online parameters. To demonstrate the versatility and adaptability of the algorithm in unknown environments, physics-based simulations on various maps are presented. Benchmark tests show the superior performance of the proposed algorithm over its previous iteration and another state-of-art method. The open-source implementation of the algorithm and the benchmark data can be found at https://github.com/RIVeR-Lab/tentabot.

Recurrent Neural Networks for Hierarchically Mapping Human-Robot Poses

To perform many critical manipulation tasks successfully, human-robot mimicking systems should not only accurately copy the position of a human hand, but its orientation as well. Deep learning methods trained from pairs of corresponding human and robot poses offer one promising approach for constructing a human-robot mapping to accomplish this. However, ignoring the spatial and temporal structure of this mapping makes learning it less effective. We propose two different hierarchical architectures that leverage the structural and temporal human-robot mapping. We partially separate the robotic manipulator's end-effector position and orientation while considering the mutual coupling effects between them. This divides the main problem---making the robot match the human's hand position and mimic its orientation accurately along an unknown trajectory---into several sub-problems. We address these using different recurrent neural networks (RNNs) with Long-Short Term Memory (LSTM) that we combine and train hierarchically based on the coupling over the aspects of the robot that each controls. We evaluate our proposed architectures using a virtual reality system to track human table tennis motions and compare with single artificial neural network (ANN) and RNN models. We compare the benefits of using deep learning neural networks with and without our architectures and find smaller errors in position and orientation, along with increased flexibility in wrist movement are obtained by our proposed architectures. Also, we propose a hybrid approach to collect the training dataset. The hybrid training dataset is collected by two approaches when the robot mimics human motions (standard learn from demonstrator LfD) and when the human mimics robot motions (LfDr). We evaluate the hybrid training dataset and show that the performance of the machine learning system trained by the hybrid training dataset is better with less error and faster training time compared to using the collected dataset using standard LfD approach.

A Quadratic Programming Approach to Manipulation in Real-Time Using Modular Robots

Motion planning in high-dimensional space is a challenging task. In order to perform dexterous manipulation in an unstructured environment, a robot with many degrees of freedom is usually necessary, which also complicates its motion planning problem. Real-time control brings about more difficulties in which robots have to maintain the stability while moving towards the target. Redundant systems are common in modular robots that consist of multiple modules and are able to transform into different configurations with respect to different needs. Different from robots with fixed geometry or configurations, the kinematics model of a modular robotic system can alter as the robot reconfigures itself, and developing a generic control and motion planning approach for such systems is difficult, especially when multiple motion goals are coupled. A new manipulation planning framework is developed in this paper. The problem is formulated as a sequential linearly constrained quadratic program (QP) that can be solved efficiently. Some constraints can be incorporated into this QP, including a novel way to approximate environment obstacles. This solution can be used directly for real-time applications or as an off-line planning tool, and it is validated and demonstrated on the CKBot and the SMORES-EP modular robot platforms.

Identifying Hazardous Shapes in the Plane

This paper explores the problem of identifying the shapes of invisible hazardous entities in R2 by a set S = {s1, s2, . . . , sk} of mobile sensors (autonomous robots). A hazardous entity, H, is a region that affects the operation of robots that either penetrate the area or come in contact with it. In this paper, we propose algorithms for searching a rectangular region for a stationary hazardous entity, where some a priori geometrical knowledge is given (e.g., edge size range), and if such an entity exists, then determine the area that it occupies. We explore entities that are convex in nature such as line segment, circles (discs), and simple convex shapes. The objectives are to minimize the distance travelled by the robots during the search phase, and to minimize the number of robots that are required to identify the region covered by the hazardous entity. The number of robots required to locate H is three or four robots when H is a line segment, two or three robots when H is a circle, and seven robots are sufficient when H is a triangle. Our results extend to n-vertex convex shapes and we show that 2n + 1 robots are sufficient to determine the coverage of H.

View Planning for Robotic Inspection of Tolerances Through Visual Tracking of Manual Surface Finishing Operations

This article presents an approach for determining suitable camera view poses for inspection of surface tolerances based on visual tracking of the tool movements performed by a skilled worker. Automated surface inspection of a workpiece adjusted by manual operations depends on manual programming of the inspecting robot, or a timeconsuming exhaustive search over the entire surface. The proposed approach is based on the assumption that the tool movements of the skilled worker coincide with the most relevant regions of the underlying surface of the workpiece, namely the parts where a manual process has been performed. The affected region is detected with a visual tracking system, which measures the motion of the tool using a low-cost RGBD-camera, a particle filter, and a CAD model of the tool. The main contribution is a scheme for selecting relevant camera view poses for inspecting the affected region using a robot equipped with a high-accuracy RGBDcamera. A principal component analysis of the tracked tool paths allows for evaluating the view poses by the Hotelling’s T-squared distribution test in order to sort and select suitable camera view poses. The approach is implemented and tested for the case where a large ship propeller blade cast in NiAl bronze is to be inspected by a robot after manual adjustments of its surface.

Real-time Driving Context Understanding using Deep Grid Net: A Granular Approach

Numerous self-driving cars algorithms rely on grid maps for motion planning, obstacles avoidance, or environment perception. Obtained from fused sensory information, the occupancy grids (OGs) are nowadays among the most popular solutions used in series production in the automotive industry. In this paper, we extend Deep Grid Net (DGN) [18], a deep learning (DL) system designed for understanding the context in which an autonomous car is driving. We consider this paper a granular approach to DGN method due to the improvements added to the original research [18]. DGN incorporates a learned driving environment representation based on OGs obtained from raw real-world Lidar data and constructed on top of the Dempster-Shafer (DS) theory. Our system is able to predict in real-time if the vehicle is driving on the highway, on county roads, inside a city, in parking lots or is stuck in a traffic jam. The predicted driving context is further used for switching between different autonomous driving strategies implemented within EB robinos, Elektrobit’s Autonomous Driving (AD) software platform. We propose a neuroevolutionary approach to search the optimal hyperparameters set of DGN. Genetic algorithms (GAs) were selected due to their demonstrated capabilities to evolve deep neural networks with improved accuracy and processing speed. The performance of the proposed deep network has been evaluated against similar competing driving context estimation classifiers.

LEO: Liquid Exploration Online

This article introduces a novel approach to the online complete- coverage path planning (CCPP) problem that is specically tailored to the needs of skid-steer tracked robots. In contrast to most of the current state-of-the-art algorithms for this task, the proposed algorithm reduces the number of turning maneuvers, which are responsible for a large part of the robot's energy consumption. Nevertheless, the approach still keeps the total distance traveled at a competitive level. The algorithm operates on a grid-based environment representation and uses a 3x3 prioritization matrix for local navigation decisions. This matrix prioritizes cardinal di- rections leading to a preference for straight motions. In case no progress can be achieved based on a local decision, global path planning is used to choose a path to the closest known unvisited cell, thereby guaranteeing completeness of the approach. In an extensive evaluation using simulation experiments, we show that the new algorithm indeed generates competi- tively short paths with largely reduced turning costs, compared to other state-of-the-art CCPP algorithms. We also illustrate its performance on a real robot.