rayvburn / humap_local_planner

Human-aware robot trajectory planner using a hybrid trajectory candidates generation and spatiotemporal cost functions
BSD 3-Clause "New" or "Revised" License
0 stars 1 forks source link

tune parameters to avoid robot being stuck in sharp corners #71

Closed rayvburn closed 1 year ago

rayvburn commented 2 years ago

Related to both costmap_converter and hubero_local_planner parameters

rayvburn commented 2 years ago

When robot is located a little further down the global plan, it produces local plan that crosses the forbidden area produced by a sharp corner:

image

rayvburn commented 2 years ago

After multiple tries with different costmap converter parameters, it seems that the only method to overcome this issue is to produce as many point obstacles as possible (here, cluster_min_pts = 1, cluster_max_pts = 2).

image

However, this produces massive computational cost and slows down planner a lot.

So, in a nutshell, trajectory generator is very sensitive to improper obstacle segmentation. This issue was previously addressed at #17.

rayvburn commented 2 years ago

After thorough debugging it seems that failures are caused by Social Force Model itself. Force grid visualization shown that robot is directed towards walls. As a remedy, another approach to trajectory generation is used ATM.