Adapt existing Visual SLAM to Atlas robot in DRCSim

I would like a SLAM solution based on vision (without using the LIDAR spindle) for an Atlas robot within the DRC Simulator ([url removed, login to view]). Any existing libraries may be used, and either Python or Java or C++ are okay (bids offering to do the work in Java will be preferred).

A simple solution is sufficient; the map does not need to be very accurate, and real walking and balancing are not necessary (fake walking is sufficient, see [url removed, login to view]). Apart from the SLAM, the code should also save the images the robot sees (once per second).

Many existing libraries / packages already implement visual SLAM. The main work consists of a) getting one of them to work with the Atlas and in the DRC environment, and b) implementing a simple wall following behavior. The learned map should be an occupancy map.

The solution code should do the following (without using LIDAR):

- Run in the existing DRC task 3 environment (which can be started after installing drcsim by calling "roslaunch drcsim_gazebo [url removed, login to view]")

- Have the robot walk around in this environment (fake walking suffices, no balancing necessary), exploring all of it, following walls and avoiding obstacles

- Maintain an approx. robot location

- Build an approx. occupancy map of the environment

- Save the current occupancy map at every second into a specified folder as an image file, including a marker indicating the robot's current position

- Save the raw image perceived by the two cameras at every second into a specified folder (i.e. two images per second, for the two cameras)

Some experience with the Robot Operating System will be necessary, preferably also with the DRC Simulator.

Please post the language you intend to develop in (C++, Python or Java) together with your bid

Ρομποτική Ανάπτυξη Λογισμικού

Ταυτότητα Εργασίας: #7874253

Σχετικά με την εργασία