Creating and storing a 2D/3D map for autonomous robot

Question

I have built an autonomous RC car that is capable of tracking its location and scanning its environment for obstacles. I am stuck however on how to store/create a map of its environment. The environment would not be fixed in size, as the robot explores it may uncover new areas. My background is on navigation, so that part and determining where obstacles are is the easy part (for me). The hard part is how to store/access the data.

Originally I was going to use a 2D array/matrix with a resolution of 2-5 cm, ie each cell is 2-5cm wide when projected onto the ground. This however could eat up a lot of memory (it is embedded so I have limited resources, 512mb of RAM right now but may decrease when prototype is finished). As well, I will eventually be moving the system to a quadcopter so I will be working in 3D. The next solution was to create an obstacle object, which would have a position. However I was unsure on how to: 1. keep track of the size and orientation of the object, ie how to tell if the object is a wall or just a table leg (cant avoid a wall, can go around a leg) and 2. how to efficiently search/tell if an obstacle is in the robots way.

I am using c++, and the integration of a camera and ultrasonic sensor to detect the environment. As mentioned, stuck on how to store the environment efficiently. I would like to keep the environment information, whether save to a file or send to a server so I can later use/access the map. Would anyone have any suggestions on what to try, or where to even begin looking, this area is all new to me?

Answer 1

You can use a R-Tree where your initial place is the root of the tree. Each time you can go somewhere, you can add the place you went in your tree. You could also fusion some neighbour parts together if possible.

From a memory point of view, you are not bounded by your array/matrix limits and you store only valuable data. It is up to you to choose what to store : obstacles or free space. Searching in that kind of structure is also efficient.

As for implementations, you could take a look a the r-tree in the boost library .

Later on, you could even create a graph based on your tree to be used by path-finding algorithms if you expect your robot to go from one point to another.