Reactive multi-robot manipulation can be of great assistance in many applications. In this article a distributed cooperation strategy for object manipulation with multiple robots in un-modelled environments is proposed to tackle workpiece transportation and dynamic obstacle avoidance simultaneously. A decentralised optimisation process will run first to generate robot pose references and then a joint-level control is responsible for tracking. Since in multi-arm manipulation systems the available workspace for each robot is more restricted, necessary internal performance optimisation is also included in the planning phase. A velocity-level optimum will be solved and then transformed to a position-level control reference so that it can be combined with many compliant interactive controllers such as impedance control. Although no torque or force sensor is required and each robot only has access to local information, when avoiding obstacles the coordination strategy can generate synchronised group motion so that internal forces are avoided. Only task-space variables will be communicated, therefore it is easy to team up different types of robots, making the system more flexible. Simulation and experiments with two redundant manipulators have been done to validate the proposed design, and data show that the online computation speed is fast.

反应式多机器人操纵在许多应用中都可以提供很大的帮助。在本文中，提出了一种在无模型环境中使用多个机器人进行对象操纵的分布式协作策略，以同时解决工件运输和动态避障问题。分散的优化过程将首先运行以生成机器人姿态参考，然后由关节级控件负责跟踪。由于在多臂操纵系统中，每个机器人的可用工作空间受到更多限制，因此在计划阶段还包括必要的内部性能优化。将求解速度级别的最佳值，然后将其转换为位置级别的控制参考值，以便可以将其与许多兼容的交互式控制器（例如阻抗控制）结合使用。尽管不需要扭矩或力传感器，并且每个机器人只能访问本地信息，但是在避开障碍物时，协调策略可以生成同步的组运动，从而避免了内力。仅会传递任务空间变量，因此很容易组合不同类型的机器人，从而使系统更加灵活。用两个冗余机械手进行了仿真和实验，以验证所提出的设计，并且数据表明在线计算速度很快。使系统更加灵活。用两个冗余机械手进行了仿真和实验，以验证所提出的设计，并且数据表明在线计算速度很快。使系统更加灵活。用两个冗余机械手进行了仿真和实验，以验证所提出的设计，并且数据表明在线计算速度很快。