Cherry Blossom Slam

Our society likes labels. People like labels. Putting things into distinct boxes reduces the complexity of life. Putting people into categories simplifies the complex communication with and about them. We also choose labels for ourselves. It is neither necessary nor impossible; conversely, it is quite arbitrary. The labels we assign are always chosen with respect to our environment and with respect to our own position.

There is a concept in autonomous automation called simultaneous localization and mapping (SLAM): An autonomous agent has to navigate in an unknown environment by using its sensors that observe the surroundings. Thus, it needs to localize itself with respect to the world and it needs to map the environment. The former step, localization, requires a map – only with a correct map it can use its sensor data to infer the current position. The latter step, mapping, requires the exact position of the agent in order to build an internal representation of the world with respect to the current sensor data. So which task should be performed first?

Generally, solutions to SLAM are careful: mapping the environment and inferring labels has to be done simultaneously and with caution. It also relies on large amounts of data. Typically, SLAM is solved via iterative updates of both, the own position, and the map with labels. In every iteration, the already gained information about the own position and environment can be used to improve the estimate of both. Early overconfidence is not advised and can result in disastrous accidents; when used in cars, and when used by people. Sometimes, the latter are not careful enough.