A study of interaction scheme and external computation resource for heterogenous autonomous robots

我们已经开发了一个集成架构多个小型割草机器人。该体系结构包括（1）一个混合方案，以提高机器人的可靠性，（2）代理框架，使机器人之间的合作，（3）调度方法的机器人，和（4）学习计划的机器人动作策略。所开发的割草机器人由基本割草单元、CPU、GPS和通信模块组成。这些问题包括（1）机器人容易在不平坦的表面上堆叠，（2）路径跟踪能力低，（3）在外部环境中通信不稳定。为了解决这些问题，我们需要为机器人设计鲁棒和高效的工作策略，包括机器人和充电站。为此，我们开发了混合方案，以提高机器人的可靠性，同时，我们开发了基于Web服务技术的代理框架，使机器人之间的合作。该框架包括Java移动的agent，机器人可以基于该框架进行协作，并证明了该方案的有效性。在机器人的动作策略学习方面，我们开发了机器人的两个学习特性。一个是用于碰撞避免，另一个是用于从卡滞情况中脱离的按需学习。仿真结果表明了这些特征的有效性。我们也在真实的机器人上实现了这些功能。对真实的机器人的实验结果与仿真结果一致。

We have developed an integrated architecture for multiple small mowing robots. The architecture includes (1) a mixture scheme to improve reliability of robots, (2) agent framework to enable cooperation among robots, (3) a scheduling methods for robots, and (4) learning scheme for robot action policy. The developed mowing robots consists of a basic mowing unit, CPU, GPS, and communication module as well.The previous work shows that small mowing robots suffers from several problems. They include (1) robots easily stack on uneven surface, (2) low path trace capability, and (3) communication is unstable in the outside environment Furthermore, robots need to be recharged when they gets lower electricity energy. Tb handle these problems, we need to device robust and efficient working strategy for robots, which includes robots and charging stations. For that purpose, we have developed mixture scheme to improve the reliability of robots.Also, we have developed agent framework based on Web service technology, which enables cooperation among robots. The framework includes Java mobile agent Robots are able to cooperate based on the framework We also show efficiency of the scheme.As far as the learning of action policy of robots is concerned, we have developed two learning features of robots. One is for collision avoidance, and the other is for on demand learning of escape from stuck situation. The simulation results shows efficiency of these features. We also implemented these feature on the real robots. The experimental results of the real robots are consistent with the simulation results.