【作者】 韩震峰；

【导师】 赵杰；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2012， 博士

【摘要】 随着科学技术的发展，在自然灾难或人为灾害发生后的废墟救灾现场将不再全部依靠人类亲自进入现场实施搜救工作。灾难搜救探测机器人的出现降低了搜救人员在灾难废墟中二次伤害的发生，具有非常重要的现实意义。限于目前的技术水平，探测机器人很难完全自主地完成所有的作业任务，遥控为主、局部自主是目前合理的控制模式。在这种模式下，操作人员会辅助机器人完成一些复杂的任务，而机器人在某些基础行为上也具有一定的自主性。本文首先研制了一台用于煤矿井下环境搜索的多节履带式移动机器人，然后针对为实现该机器人具有一定程度的局部自主能力所涉及到的关键技术进行了研究，包括基于超声和单目视觉系统的障碍物及自身运动的感知研究、多节履带式机器人的定位研究以及该类型机器人的一些基本行为控制方法和控制策略研究。针对煤矿井下的特殊环境，设计了一种多节履带式移动机器人。该机器人采用正压防爆设计，四周包覆履带，具有较强的环境适应能力，能够在翻滚时继续进行搜索任务；机器人为四节串联，串联关节机构具有俯仰和偏航两个主动自由度，这种设计使机器人具有较强的运动能力，能够攀爬较高的障碍物。机器人上配置了超声传感器、视觉传感器、姿态传感器、霍尔传感器等多种传感器，为实现机器人的局部自主和导航提供了硬件条件。针对该机器人多节的结构特点，设计了基于CANOpen总线协议的分布式控制系统，在传感器集中配置的两节中开发了用于传感器数据处理的MicroCANOpen子节点；开发了机器人单目视觉系统用于视觉导航应用；设计了基于Zigbee技术的无线通讯节点用于与上位机通讯，通讯链上设计了数据中继节点，可以保证在远距离或有遮挡环境下的通讯可靠性；为控制系统设计了混合式控制体系结构框架，该框架为完成搜索任务目标，把感知、建模、规划、决策和行动等多个功能模块进行了有机、高效的结合；设计了针对煤矿井下环境探测的传感器系统，可以探测甲烷、CO、温度以及湿度等重要环境参数；搜索环境的高度非结构化与未知性决定了机器人行为的复杂性，为设计与快速验证这些行为建立了机器人仿真系统，通过该系统能够对机器人的行为特性进行快速仿真分析，为实体机器人的控制算法研究提供依据及仿真验证平台。该机器人系统经实验测试在非结构化环境中具有良好的搜索性能。受机器人结构的限制，机器人只装配了单目视觉系统。视觉传感器提供了机器人上装配的传感器中最为丰富的信息量，是机器人完成局部自主与导航不可缺少的信息源。由于搜索场所通常是光线昏暗的环境，视觉传感器采用具有红外功能的CCD摄像头，并针对红外图像噪声大亮度低的特点，建立了基于Contourlet变换的图像预处理算法，有效地消除了噪声，并对图像亮度进行了平衡；提出了基于近前下方地平假设的机器人运动估计算法，该算法通过对近前下方地面强角点光流的计算对机器人的运动状态进行有效估计，为机器人定位提供了一种信息源；结合机器人的多节结构特点，基于超声测距及对视觉图像的光流场发散性分析对视景范围内的障碍物进行检测，建立视景障碍物图，为机器人避障导航提供依据。航位推算，也称为相对定位，是机器人导航的基础。通过机器人上装配的微惯性传感器建立捷联惯性导航系统，用于短时间内的姿态参考；建立了多节履带式机器人的稳态运动模型，对各种转向模式进行分析，提出了一种基于轨迹曲率与侧偏角的航向估计方法，在此基础上建立了多节履带式机器人的运动模型；通过卡尔曼滤波器对捷联惯性姿态参考系统进行信息融合，通过无迹卡尔曼滤波器对视觉运动估计系统进行信息融合，实现机器人的相对定位。行为级目标自主是任务级目标自主的基础，是机器人实现局部自主的必要条件，机器人的行为便是完成这些目标的相对独立的功能模块。对多节履带式机器人的基本行为进行了分析，建立了针对多节履带式机器人的用于转向及航向控制的模糊控制器；对沿墙跟踪行为与避障行为策略进行了研究；基于行为向导的方式对机器人越障进行了研究；建立了搜索机器人示范应用实验系统，进行了煤矿井下现场应用实验，对机器人的运动性能与控制系统性能进行现场验证测试。实验结果表明所研制的多节履带式探测机器人在搜索探测方面具有较大优势，基于本文理论方法能够很好地辅助该机器人完成相关搜索任务。更多还原

【Abstract】 With the development of technology, the search and rescue work in the disasters willnot only depend on human beings. The search and rescue robot prevents the injury of therescue people. The study of search and rescue robot that can replace human beings andgo to the disaster places possessing very important realistic meaning. Due to the limitedlevel of technology, the robot cannot finish the entire job by itself. The strategy of mainlydepending on remote control and partly depending on self-guidance is preferred. Underthis strategy, the operator can help the robot to finish the complex and difficult tasks. Andthe robot has some low-level autonomy. First, a serpentine tracked robot for underminesearching was develped. Then some key technologies for realizing its low-levelautonomy were studied. The first was obstacle detecting and self-motion detecting basedon the ultrasonic sensor and onocular vision. The second was location of serpentinetracked robot and the last was the control method and strategy.Facing the highly unstructured environment after disasters, this thesis designed aserpentine tracked robot. This robot is omni tracked and has strong ability to adapt theenvironment. It can keep work and even roll over which would definitely happen whileworking. The robot has four segments. The joint between two segments has yawing andpitching degrees. The design makes the robot have strong motion capability and canclimb high obstacles.This robot is equipped with ultrasonic distance sensors, the visionsensor, attitude sensor, hall sensors and other sensors. The navigation of robot can bebased on these sensors. Considering the structure of the robot, we constructed thedistributed control system based on the CANOpen bus. The MicroCANOpen nodes fordealing with the sensor information were developed and these two nodes were fixed inthe head segment and the tail segment. A monocular system for vision navigation wasdeveloped and studied. The wireless communication nodes and relay nodes based on thetechnology of Zigbee were developed. These nodes can keep reliable communicationwhile the distance is far or there are obstacles. A framework and architecture for controlsystem were also designed. This framework realizes seamless integration of sensor,model, plan, decision and motion. The sensor system for undermine environment wasdeveloped. The system can detect CH4, CO, temperature, dampness and other necessaryenvironment parameters. The high unstructured and unpredictable environment makessome motion of the robot very complex. To design the motions fast and to test thesemotions, a simulation system was set up. By the simulation on this system, we can getthe behavioral characteristics and result rapidly. And the result can offer reference andgist the remote control of the robot. The robot was proven to have good searching ability by experiments.Because of the limition of structure, the monocular vision system was equipted. Thevision sensor provides the richest information about the environment. The visioninformation is indispensable for navigation. Because the environment of rescue is oftendark, the vision sensor is a CCD camera with infrared vision. To resolve the problem ofbig noises in the infrared vision, the preprocessing algorithms based on the Contourlettransformation was studied. This algorithm has good results for image denoising andlight balance. A method of motion estimation based on the hypothesis of flat ground offront downward region is suggested. The method estimates the robot motion by pickingup the strong corners in the region and calculates the optical flow. And this methodprovides another information for location By analyzing the divergent of optical flow,obstacles in the vision can be detected. The obstacles diagram can be set up which willhelp the robot’s strategy of obstacle avoiding.Location is the base of robot navigation. The strap down inertial navigation system(SINS) is set up which can be attitude reference in short time. The stable movementmodel for serpentine tracked robot is set up. A linearization method of track curvatureand deviating angle is suggested based on the analysis of turning modes. Odometer forserpentine tracked robot is proposed based on the previous study. The Kalman filter forinformation fusion between the odometer and SINS is set up. And the UKF for theinformation fusion between the odometer and vision information is proposed too. Thesestudies realize the location of the robot.System level objective autonomy is the base of task level objective autonomy. It isthe necessary condition of partly autonomous navigation. The motions of the robot arerelatively independent function models to get to these targets. The typical motion forserpentine tracked robot is studied. A fuzzy controller for turning direction motion is setup. The wall follow motion and the obstacle avoiding motion are realized. In addition,the simulations or experiments about these typical motions are done.The experimentsystem for the search and rescue robot in coalmine was set up. The search and rescueexperiment in fields were done and this experiment verified the integral performance. Allof the experiments show that the serpentine tracked robot has the special advantage insearch and rescue jobs and verify the methods and algorithms in this thesis.更多还原