Evolving hierarchical robot behaviours

Inspired by the work of Brooks, many researchers involved in programming robots have turned to the behaviour-based approach. At present, the behaviours are designed by hand and hard-wired into the architecture. The work presented in this paper looks at using an evolutionary algorithm approach (based on the genetic algorithm) to construct behaviours. Building from well-defined primitive behaviours, hierarchies can be evolved to produce more complex behaviour. The behaviours in the evolutionary system are tested in simulation, but the best are then tested on a mobile robot for grounding in the real world. This allows the evolutionary process to rapidly drive the development of the behaviours using simulation while also ensuring their suitability in the real world. In the paper we show how this evolutionary process evolves practical hierarchical behaviours for the detection of a goal object in a series of mazes.     

Towards a human–robot symbiotic system

Partnership between a person and a robot could be simplified if the robot were intelligent enough to understand human intentions and perform accordingly. During the last decade, we have been developing such an intelligent robot called ISAC. Originally, ISAC was designed to assist the physically disabled, but gradually became a test bed for more robust human–robot teaming (see http://eecs.vanderbilt.edu/CIS/). In this paper, we will describe a framework for human–robot interaction, a multi-agent based robot control architecture, and short- and long-term memory structures for the robot brain. Two applications will illustrate how ISAC interacts with the human.     

A tree architecture with hierarchical data processing on a sensor-rich hexapod robot

This paper presents the hardware and software architecture of Golem, a hexapod robot designed as a flexible, scalable, general purpose development and experimenting tool targeted to academia, industry, and defense environments. The system is technologically innovative in its architecture, performance, size and integration, and is industrially promising in its filling the gap between low-performance commercial solutions and costly application-specific proprietary solutions.     

Towards human motion capture from a camera mounted on a mobile robot

This article describes a multiple feature data fusion applied to a particle filter for marker-less human motion capture (HMC) by using a single camera devoted to an assistant mobile robot. Particle filters have proved to be well suited to this robotic context. Like numerous approaches, the principle relies on the projection of the model's silhouette of the tracked human limbs and appearance features located on the model surface, to validate the particles (associated configurations) which correspond to the best model-to-image fits. Our particle filter based HMC system is improved and extended in two ways. First, our estimation process is based on the so-called AUXILIARY scheme which has been surprisingly seldom exploited for tracking purpose. This scheme is shown to outperform conventional particle filters as it limits drastically the well-known burst in term of particles when considering high dimensional state-space. The second line of investigation concerns data fusion. Data fusion is considered both in the importance and measurement functions with some degree of adaptability depending on the current human posture and the environmental context encountered by the robot. Implementation and experiments on indoor sequences acquired by an assistant mobile robot highlight the relevance and versatility of our HMC system. Extensions are finally discussed.     

Monge mapping using hierarchical NURBS

Texture mapping is an efficient and effective tool in computer graphics and animation. While computationally very cost-effective, texture mapping may produce non-realistic appearances of shapes in 3D environment, especially when viewing closely. To improve the realism of 3D modeling, bump mapping technique is developed to add details with the 3D models on top of texture mapping. Bump mapping, however, offers only simple and visual enhancement. Displacement mapping technique can further improve the localized detail of geometry. In this paper, Monge mapping technique is developed for detail and local shape modification of NURBS represented geometry in a 3D environment. Based on multiresolution and refinement schemes, Hierarchical NURBS (H-NURBS) is first investigated to design a mechanism for the purpose of carrying localized geometric information. Monge mapping on H-NURBS patch can be easily performed via simple cut-and-paste operation. Parametric control of the local shapes is developed to facilitate easier and better 3D local modeling.     

Towards development of a slider-crank centered self-propelled dolphin robot

This paper is devoted to the mechatronic design and implementation of a dolphin-like robot capable of fast swimming. In the context of multiple coordinated control surfaces, a set of serially connected flapping modules is responsible for dorsoventral oscillations, an internal moving slider for pitch control and a yaw joint for lateral turns. To improve the swimming speed, an updated modular slider-crank-based flapping mechanism that fully capitalizes on the continuous high-speed rotation of the DC motor is proposed and constructed. With the proposed mechanisms, the resulting dolphin robot achieved a high level of propulsive speed, largely illustrating the validity of the present design scheme.     

Efficient mapping algorithms for a class of hierarchical systems

Proposes techniques for mapping application algorithms onto a class of hierarchically structured parallel computing systems. Multiprocessors of this type are capable of efficiently solving a variety of scientific problems because they can efficiently implement both local and global operations for data in a two-dimensional array format. Among the set of candidate application domains, low-level and intermediate-level image processing and computer vision (IPCV) are characterized by high-performance requirements. Emphasis is given to IPCV algorithms. The importance of the mapping techniques stems from the fact that the current technology cannot be used to build cost-effective and efficient systems composed of very large numbers of processors, so the performance of various systems of lower cost should be investigated. Both analytical and simulation results prove the effectiveness and efficiency of the proposed mapping techniques     

Intelligent hierarchical robot control for sewing fabrics

In this paper an intelligent hierarchical controller for the robotized sewing of two plies of fabrics is presented. The proposed system is based on the concept: fabric properties estimation – tensional force determination – sewing – adaptation. A new methodology for integrating the tensile test of fabrics into the robotic sewing station using the sewing machine is presented. The output of this test is the estimation of the fabrics extensibility, which is fed to the next level of decision making to determine the appropriate fabric tensional force that should be applied during the sewing process. Computational intelligence methods (fuzzy logic and neural networks) have been used throughout the hierarchical structure of the controller. The present research is focused on the concept of using qualitative properties of the fabrics and the processing of qualitative and quantitative knowledge in different levels of the introduced hierarchical system. The proposed system is flexible, adaptable and robust enough to sew a wide range of unknown double ply of fabrics as it is shown by the test results. It has also the capability of on-line and endless training in order to be able to respond, handle and sew new types of fabrics. Seams that are produced by the robot and a human operator for joining two pieces of fabrics are presented and compared.     

Real-time hierarchical POMDPs for autonomous robot navigation

This paper proposes a new hierarchical formulation of POMDPs for autonomous robot navigation that can be solved in real-time, and is memory efficient. It will be referred to in this paper as the Robot Navigation–Hierarchical POMDP (RN-HPOMDP). The RN-HPOMDP is utilized as a unified framework for autonomous robot navigation in dynamic environments. As such, it is used for localization, planning and local obstacle avoidance. Hence, the RN-HPOMDP decides at each time step the actions the robot should execute, without the intervention of any other external module for obstacle avoidance or localization. Our approach employs state space and action space hierarchy, and can effectively model large environments at a fine resolution. Finally, the notion of the reference POMDP is introduced. The latter holds all the information regarding motion and sensor uncertainty, which makes the proposed hierarchical structure memory efficient and enables fast learning. The RN-HPOMDP has been experimentally validated in real dynamic environments.     

Sonar mapping of a mobile robot considering position uncertainty

This paper suggests a new sonar mapping method considering the position uncertainty of a mobile robot. Sonar mapping is used for recognizing the unknown environment for a mobile robot during navigation. Usually accumulated position error of a mobile robot causes considerable deterioration of the quality of a constructed map. In this paper, therefore, a new Bayesian probability map construction method is proposed, which considers estimation of the position error of a mobile robot. In this method, we applied approximation transformation theory to estimate the position uncertainty of a real mobile robot, and introduced cell ordering uncertainty caused by the position uncertainty of a robot in cell-based map updating. Through simulation we showed the effect of a robot's position uncertainty on the quality of a reconstructed map. Also, the developed methods were implemented on a real mobile robot, AMROYS-II, which was built in our laboratory and shown to be useful enough in a real environment.     

基于点线相合的机器人增量式地图构建

提出基于测量数据点和已构建地图线的最佳相合性增量式构建未知环境地图的方法,将机器人地图构建过程分解为局部地图构建、机器人位姿估计和地图合并3个循环步骤,在局部地图构建中,采用哈夫变换拟合、同线性判断和最小二乘拟合相结合的方法从测量数据点中拟合得到局部线段集合,在位姿估计时,首先利用点线匹配寻找测量数据和已构建地图之间的匹配关系,然后通过去除不当匹配和引入加权矩阵来减小测量误差和已构建地图中的不确定性对位姿估计的影响,最后利用加权最小二乘法估计机器人的位姿,使得匹配部分达到最佳相合,同时提出虚拟线和虚拟点的方法解决由伪相合条件所引起的错误位姿估计问题,实验结果证明了算法的有效性和鲁棒性,适于构建室内环境地图。     

Management system for distributed and hierarchical robot groups

This paper introduces a management system for robot groups with distributed and hierarchical architecture. In order to manage cooperation tasks of a great many robots, we put management systems on each work sub-area. A management system on a certain sub-area divides a task given by the upper sub-area into sub-tasks and gives the lower sub-areas those sub-tasks. With this architecture, we obtain distributed and hierarchical management system for the whole working area. We produced a simulation system and showed an example of task division and execution. We also produced a demonstration system.     

Towards efficient flight: insights on proper morphing-wing modulation in a bat-like robot

In this article, we address the question of how the flight efficiency of Micro Aerial Vehicles with variable wing geometry can be inspired by the biomechanics of bats. We use a bat-like drone with highly articulated wings using shape memory alloys (SMA) as artificial muscle-like actuators. The possibility of actively changing the wing shape by controlling the SMA actuators, let us study the effects of different wing modulation patterns on lift generation, drag reduction, and the energy cost of a wingbeat cycle. To this purpose, we present an energy-model for estimating the energy cost required by the wings during a wingbeat cycle, using experimental aerodynamic and inertial force data as inputs to the energy-model. Results allowed us determining that faster contraction of the wings during the upstroke, and slower extension during the downstroke enables to reduce the energy cost of flapping in our prototype.     

Towards a unified visual framework in a binocular active robot vision system

This paper presents the results of an investigation and pilot study into an active binocular vision system that combines binocular vergence, object recognition and attention control in a unified framework. The prototype developed is capable of identifying, targeting, verging on and recognising objects in a cluttered scene without the need for calibration or other knowledge of the camera geometry. This is achieved by implementing all image analysis in a symbolic space without creating explicit pixel-space maps. The system structure is based on the ‘searchlight metaphor’ of biological systems. We present results of an investigation that yield a maximum vergence error of ～6.5 pixels, while ～85% of known objects were recognised in five different cluttered scenes. Finally a ‘stepping-stone’ visual search strategy was demonstrated, taking a total of 40 saccades to find two known objects in the workspace, neither of which appeared simultaneously within the field of view resulting from any individual saccade.     

基于分层Option的仿人机器人相似性关键姿势转换

针对运动捕获系统获取的人体运动轨迹固定、难以实现仿人机器人关键姿势转换问题,提出了一种基于分层Option学习的仿人机器人关键姿势相似性转换方法。构建多级关键姿势树状结构,从关节相似差异、时刻整体相似差异、周期整体相似差异等角度描述了关键姿势差异,引入分层强化Option学习方法,建立关键姿势与Option行为集,由关键姿势差异的累计奖励将SMDP-Q方法逼近最优Option值函数,实现了关键姿势的转换。实验验证了方法的有效性。     

基于分层深度强化学习的移动机器人导航方法

针对现有基于深度强化学习(deep reinforcement learning, DRL)的分层导航方法在包含长廊、死角等结构的复杂环境下导航效果不佳的问题,提出一种基于option-based分层深度强化学习(hierarchical deep reinforcement learning, HDRL)的移动机器人导航方法.该方法的模型框架分为高层和低层两部分,其中低层的避障和目标驱动控制模型分别实现避障和目标接近两种行为策略,高层的行为选择模型可自动学习稳定、可靠的行为选择策略,从而有效避免对人为设计调控规则的依赖.此外,所提出方法通过对避障控制模型进行优化训练,使学习到的避障策略更加适用于复杂环境下的导航任务.在与现有DRL方法的对比实验中,所提出方法在全部仿真测试环境中均取得最高的导航成功率,同时在其他指标上也具有整体优势,表明所提出方法可有效解决复杂环境下导航效果不佳的问题,且具有较强的泛化能力.此外,真实环境下的测试进一步验证了所提出方法的潜在应用价值.     

基于二维激光的机器人室内构图

建立了2D激光扫描匹配数据的构图方案,使用GMapping(一种高效的Rao-Blackwellized粒子滤波器,从激光数据学习网格地图)通过改变内部扫描配准算法,实现录制数据(原始2D扫描数据)经过激光配准算法后生成相对运动文件作为tf,用以提高地图生成和激光扫描配准的精度。实验基于机器人操作系统(ROS),开源的GMapping软件包进行同时定位和地图构建(SLAM),通过RVIZ(用于ROS的3D可视化工具)控制机器人运动。该方案可应用于移动机器人的导航、构图与2D重建、同步定位与地图构建(SLAM)等领域。     

基于随机行走的群机器人二维地图构建

机器人在未知环境自主探索时,需要快速准确地获取环境地图信息。针对高效探索和未知环境的地图构建问题,将随机行走算法应用于群机器人的探索中,机器人模拟布朗运动,对搜索区域建图。然后,改进了布朗运动算法,通过设置机器人随机行走时的最大旋转角度,来避免机器人重复性地搜索一个区域,使机器人在相同时间内探索更多的区域,提高机器人的搜索效率。最后,通过搭载激光雷达的多个移动机器人进行了仿真实验,实验分析了最大转角增量、机器人数量以及机器人运动步数对搜索区域的影响。     

River mapping from a flying robot: state estimation, river detection, and obstacle mapping

Accurately mapping the course and vegetation along a river is challenging, since overhanging trees block GPS at ground level and occlude the shore line when viewed from higher altitudes. We present a multimodal perception system for the active exploration and mapping of a river from a small rotorcraft. We describe three key components that use computer vision, laser scanning, inertial sensing and intermittant GPS to estimate the motion of the rotorcraft, detect the river without a prior map, and create a 3D map of the riverine environment. Our hardware and software approach is cognizant of the need to perform multi-kilometer missions below tree level with size, weight and power constraints. We present experimental results along a 2?km loop of river using a surrogate perception payload. Overall we can build an accurate 3D obstacle map and a 2D map of the river course and width from light onboard sensing.