This paper describes a method to compare and evaluate mobile robot algorithms for long-term deployment in changing environments. Typically, the long-term performance of state estimation algorithms for mobile robots is evaluated using pre-recorded sensory datasets. However such datasets are not suitable for evaluating decision-making and control algorithms where the behaviour of the robot will be different in every trial. Simulation allows to overcome this issue and while it ensures repeatability of experiments, the development of 3D simulations for an extended period of time is a costly exercise.udIn our approach long-term datasets comprising high-level tracks of dynamic entities such as people and furniture are recorded by ambient sensors placed in a real environment. The high-level tracks are then used to parameterise a 3D simulation containing its own geometric models of the dynamic entities and the background scene. This simulation, which is based on actual human activities, can then be used to benchmark and validate algorithms for long-term operation of mobile robots.
展开▼
