A Test-bed Architecture And Trajectory-analysis-based Performance Evaluation Method For Un-manned Land Vehicle

Constructing test-bed architecture and evaluation approach remains a fundamental yet challenging task for Unman Land Vehicle (ULV) since recent years witness great techno-logical progress in field of mobile robot. As an essential component of ULV, the testing and evaluation module is central to goal of navigation capacity in that it provide feedback and data analysis with both qualitative and quantitative paradigms. Therefore, this thesis discusses the test-bed architecture and a novel trajectory evaluation method based largely on MDP model.First of all, the prior arts of ULV testing and evaluation technology are surveyed. Towards better performance and effectiveness, several challenges are indicated including systematic definition, testing norms, and a quantification method. Also, a few principles are proposed, for example real-time demand, minimum invasion, and least interference.Based on specific robot design, a reliable test-bed is introduced. This test-bed consists of message dispatchers as well as database manipulator. It functions as an online monitor, a real-time data collector, and a playback tool. With convenient and scalable configuration, testing data can be easily stored, displayed and fetched from database.To evaluate the navigation performance, this thesis also put forward a novel trajectory analysis approach, which assess and compares the navigation performances by reward set-tings and feature expectations underlying the Markov Decision Process framework. Assum-ing that the reward function is a linear combination of known features, this approach eval-uates reward settings by approximating the driving styles of different Land Autonomous Navigation Systems. On this basis, reward settings are used to simulate the feature ex-pectations of navigation trajectories in a standard land navigation environment called the sandbox. Experimental results show the validity of this approach in both qualitative and quantitative performance evaluations. At the last part of this thesis, contributions of this work are summarized and further researches are also suggested.