毕业生知识库

双足机器人（人形机器人）是机器人领域研究的制高点，将引领下一代技术发展和产业升级。双足机器人的真正用途是自主前往不安全或人类难以到达的地点，代替人类完成繁重、重复性的任务。目前一流双足机器人已初步具备稳定行走能力、灵巧作业能力和多模态运动技能。为进一步提升双足机器人的长续航和复杂环境高通过性等自主运动能力，以应对现实世界的复杂未知环境，本文将深入研究双足机器人安全高效自主导航与避障方法，尝试解决机器人有限传感范围且存在未知障碍物的条件下，综合考虑机器人本体安全、在线决策规划效率、高通过性和能耗水平的双足机器人局部质心运动规划、落脚点规划问题。本文主要工作如下：

1. 针对双足机器人自主行走的安全保障问题，基于机器人前向运动可达集，提出一种双足机器人安全自主运动可达集（质心局部运动可达集和落脚点可达集）构建方法。首先，通过双足机器人连杆简化模型，采用可达性分析，构建双足机器人质心的局部自主运动可达集；其次，提出一种双足机器人中间等效模型，建立质心到落脚点的映射，构建落脚点可达集。离线构建的可达集，作为在线规划的先验知识用于快速获得安全轨迹集合，既保障了自主运动时的安全性，又减少了在线规划时间，提高了在线决策规划的效率。

2. 针对双足机器人在复杂环境中自主导航与避障问题，提出一种基于安全自主运动可达集的质心轨迹和落脚点在线自主决策与规划算法。首先，基于模糊逻辑区分环境复杂程度，选择对应的算法进行质心轨迹优化；其次，针对复杂狭窄的环境离线训练深度强化学习模型，在线结合环境信息和机器人本体状态信息，实时生成目标函数进行质心轨迹优化；最后，通过双足机器人中间等效模型，进行落脚点规划。该算法通过在线实时生成目标函数的形式，降低了双足机器人自主运动能耗，提升了双足机器人在复杂环境中的运动速度和可通过率。

3. 搭建物理引擎仿真环境，采用现有成熟的双足机器人模型，开展双足机器人自主行走实验。首先，基于Matlab和ROS分别搭建了物理引擎仿真环境，用于所提出算法的有效性验证和定量评价；其次，设计了一种随机地图生成方法，制定了随机地图复杂度评价指标，进而定义一种双足机器人自主运动定量评价指标；最后，开展了双足机器人自主导航与避障综合实验验证。基于生成的随机地图数据集开展双足机器人自主行走实验，结果表明本文提出算法相较于现有算法，通过率提升了10.6%，通过时间平均降低8.94%，能耗指标平均降低了27.27%，能通过地图的复杂度平均提升了24.94%，自主运动能力评价指标提升了19.26%。基于现实世界复杂地形地图数据开展双足机器人自主行走实验，进行了复杂地形地图下的多次双足机器人自主行走实验，结果表明本文提出算法相较于现有算法，通过时间平均降低11.70%，能耗指标平均降低了14.82%，进一步验证了提出算法的有效性。研究成果将为实验室后续实物双足机器人自主行走实验、机器人自主运动能力提升奠定了坚实的理论与技术基础。

Biped robots (humanoid robots) are the research frontier of robotics, which will lead the next generation of technological development and industrial upgrading. The true application of biped robots is to autonomously go to unsafe or hard-to-reach places, and replace humans in performing heavy and repetitive tasks. Currently, leading biped robots possess preliminary stable walking ability, dexterous operation ability, and multimodal motion skills. In order to further enhance the autonomous motion capabilities of biped robots, such as long endurance and high traversability in complex environments, to cope with the complex unknown environments of the real world, this paper deeply studies the safe and efficient autonomous navigation and obstacle avoidance methods for biped robots. This article attempts to solve the problem of local centroid motion planning and footstep planning for biped robots while taking into account the robot's own safety, online decision-making efficiency, high traversability, and energy consumption levels, under the conditions of limited sensing range and unknown obstacles. The main work of this paper is as follows:

1. Aiming at the safety guarantee of biped robot autonomous walking, based on the robot forward reachable set, this paper proposes a method to construct a safe autonomous motion reachable set (centroid local motion reachable set and footstep reachable set) for biped robot. Firstly, through simplified model of biped robot and using reachability analysis, the local autonomous motion reachable set of the biped robot's center of mass is constructed. Secondly, a biped robot intermediate equivalent model is proposed to establish the mapping from the center of mass to the footstep and construct the footstep reachable set. The offline constructed reachable sets serve as prior knowledge for online planning, and the safe trajectory set is obtained after interacting with environmental obstacles to ensure the safety of biped robot online autonomous movement. This method reduces the calculation time of online planning by constructing reachable sets offline, improving the efficiency of decision planning.

2. To solve the problem of autonomous navigation and obstacle avoidance for biped robot in complex environments, this paper proposes an online autonomous decision-making and planning algorithm for centroid trajectory and footstep based on safe autonomous motion reachable set. Firstly, based on fuzzy logic to distinguish the complexity of the environment, the corresponding algorithm is selected to optimize the center of mass trajectory. Secondly, a deep reinforcement learning model is offline trained for complex narrow environments, and the target function is generated in real-time by combining environmental information and robot state information for center of mass trajectory optimization. Finally, footstep planning is performed using the intermediate equivalent model of the biped robot. This algorithm reduces the energy consumption of biped robot autonomous motion, improves the speed and traversability of biped robot to pass through complex environments by generating the target function online in real-time.

3. A physical engine simulation environment is built, and experiments on biped robot autonomous walking are conducted using the existing biped robot model. Firstly, two physical engine simulation environments are separately built using Matlab and ROS for the validation and quantitative evaluation of the proposed algorithms. Secondly, a random map generation method is designed, and a complexity evaluation index for random maps is established. Furthermore, a quantitative evaluation index for the autonomous motion capability of biped robots is defined. Finally, comprehensive experiments on autonomous navigation and obstacle avoidance of biped robots are conducted. The biped robot autonomous walking experiments are carried out based on the generated random map data set, and the results show that compared with existing algorithms, the pass-through rate is increased by 10.6%, the average passing time is reduced by 8.94%, the energy consumption index is reduced by 27.27%, the complexity index of the maps that can be passed is increased by 24.94%, and the evaluation index of autonomous motion capability is increased by 19.26%. Multiple bipedal robot autonomous walking experiments are conducted on real-world complex terrain maps, and the results show that compared with existing algorithms, the average passing time is reduced by 11.70%, and the energy consumption index is reduced by 14.82%, further verifying the effectiveness of the proposed algorithm. The research results will lay a solid theoretical and technical foundation for the follow-up physical biped robot autonomous walking experiment in the laboratory and the improvement of robot autonomous motion ability.