适应复杂环境的模块化变形机器人关键技术

The core concept of Modular Self-Reconfigurable (MSR) robots lies in the adaptability to complex environment through reconfiguration of morphology and functionality, which is unaccessible to traditional fixed-type robot. This research will provide absolutely new scientific ideas and technical approaches for improving the environmental adaptability of future robots. This project focuses on structure adjustment and function emergence of MSR robots in complex environments by introducing the self-organization and evolution of natural life. Four central and basic science issues in the development of MSR robots are solved, including: “mechanism for decentralized perception, transmission and response mechanism of environment information”, “methods for environment driven self-evolution and process control”, “efficient computation for motion planning & evolution of MSR robots”, “mechanism for emergence of adaption in both self-organization patters and locomotion strategies during interacting with environment”. More work will also be carried out on those basic theories and key technologies about MSR robots, such as environment perception, structure reconfiguration, coordinated motion, autonomous adaption. To verify the efficiency of proposed mechanisms and technologies, multiple simulations on the designed dynamical virtual platform and experiments on mechanical robots will be did by exploring unknown environments. The breakthrough of this project not only provide new research approaches for self-organization morphology emergence and distributed cooperative control of modular robot, but also can improve the engineering technique for robots adaption to complex environments.

模块化变形机器人的核心理念是通过结构重组、形态重塑、功能再生实现传统的固定构型机器人所无法达到的复杂环境适应能力，为从根本上提高未来机器人的环境适应能力、任务执行能力提供了全新的科学思路与技术途径。本项目尝试从生命自组织演化与发展进化的角度看待自重构机器人形态与功能的形成，围绕“分布式架构下的环境信息感知、传递与响应机制”、“由环境引导的机器人自组织形态演化与过程调控”、“模块化机器人多模式运动能力高效进化计算方法”、“适应复杂环境的形态重塑与适应性行为产生机理”四个基础科学问题，开展模块化机器人环境感知、形态重构、协调运动、自主行为等方面的基础理论与关键技术研究，建立面向复杂环境的机器人虚拟仿真与实验平台并完成相关理论及技术的实验验证，为模块化变形机器人的发展提供重要的理论参考与工程技术基础。

模块化变形机器人相比于传统固定机器人具备“可重构，可替换；一机多用，功能复用”的特点，能够通过改变自身的构型适应复杂的环境与任务需求，在诸多作业场景下具有应用潜力。本项目针对模块化变形机器人系统在感知，通讯，变形控制方面涌现出的特色科学问题展开了深入研究，通过借鉴生物分布式环境感知、自组织形态生长、从低级到高级的生命体进化等自然界规律，制定了感知-共享-重构-进化-智能的研究路线，提出了包括“基于无，弱，半监督的移动式双目立体视觉三维场景重建”，“基于冯-诺依曼邻居空间模型的信息梯度扩散”，“基于分形理论的混合式重构变形策略”，“基于行为稀疏度的控制器参数进化”，“基于内在调控强化学习机制的递归神经网络”等一系列技术理论框架。此外，开发了模块化变形机器人高效动力学仿真软件，研制了基于电-永磁铁连接与通讯机制的新型高集成度机器人模块。本项目为面向复杂环境的大规模自重构机器人系统的高效自动规划与控制提供了新的途径，为模块化机器人发展提供重要的理论参考与工程技术基础，同时也为机器人适应复杂环境提供一种新的科学思路。