**5.1. Evaluation method**

To visualise the evolution in computation of search-based algorithms, we compare the replanning computation of D\* Lite, Anytime Dynamic A\* and D\* Lite with Reset. The purpose of the comparison is to demonstrate the performance improvements of D\* Lite variants in order to apply on robot that operates in complex and dynamic environment. However, since the planning time depends on the implementation and machine configuration, we therefore choose the amount of cell expansion in each replanning iteration of search-based algorithm to be standard performance measurement of the mentioned algorithms. This method is independent on machine specifics and actual implementation and therefore firmly accurately shows the enhancement of this evaluation. The path solution ratio between AD\* with different *ε* suboptimal bound and optimal path of other algorithms is also measured to visualise the trade-off between optimality and computation.

The experiments are conducted on our 2D simulation engine. The state space is a 2D grid cell with uniform resolution [23]. The conceptual robot in this simulation has two-cell-unit range and its own known grid map to detect environment changes (unblocked cell to blocked cell and vice versa) as it moves along the initial path (see **Figure 15**).
