Starting from recent papers novelties, we have extended the previously proposed methodologies for navigation in complex uneven terrains [ Ferrari et al. (2019) ] by developing a footstep planner based on a randomized algorithm for an autonomous humanoid robot in a multi-floor environment. This was implemented by leveraging a multi-level surface map [ Triebel et al. (2006) ].
Tests have been carried out on three different environments, each of them with peculiar characteristics, in order to assess the correct behaviour.
Finally, the results were validated with simulations on CoppelliaSim using a gait generator based on IS-MCP for computing the CoM trajectory.
pip3 install -r requirements.txt
Translate a scene in Coppelia into a file containing all the useful information about the objects in the scene.
It is available here.
It is an extrension of the popular Elevation Map that allows to represent environments composed by multiple layers.
It exploits the information taken from the scene to build a map that stores in its cells a list of tuples, one for each object above the considered cell. Each tuple contains the height of the upper surface and a depth value representing the thickness of the object belonging to the considered cell.
RRT to search for a path to the Goal region.
More examples can be found in the outputs folder, each with its plot.
To launch the planner with default settings:
./main.py --world data/world_of_stairs.json --time-max 10000 --resolution 0.02
- world: path to the json containing all the info about the objects in the scene
- time-max: number of iterations of the RRT
- resolution: size for the world discretization
Use the help option for more details.
To convert the .tsv
footstep plan in order to be published as ROS topic:
./convert-tsv-footstep-plan.py --input TSV_FILE_PATH --output TXT_FILE_PATH