Multi-granularity Navigation for Self Service Moving

Abstract

The demand on self service moving tools like smart wheelchair become urgent with the development of society. Traditional moving facilities performs poorly in indoor environments and is unable to do fine-grained navigation in outdoor environments with GPS locators. Base on simultaneous localization and mapping with heterogeneous sensors and dynamic navigation with threat degree, we introduced a multi-granularity navigation approach for self service moving tools. Visual Inertial Odometry measurements are integrated with readings from GPS for target orientation and generates probabilistic octree represented 3D maps that fitted with real environment, providing dynamic probabilistic octree navigation for self service moving tools. This approach is able to correct visual odometry errors with inertial and GPS readings. The multi-granularity environment representation fused with probabilistic octree has taken sensor characteristics and mapping accuracy into concern and is able to achieve autonomous navigation without any prior knowledge. Experiments demonstrate the effectiveness in minimizing trajectory error under comprehensive material and luminance conditions. This approach also provides theoretical principle for research and development in self service moving facilities.