You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When running pick_ik with robot models that have continuous (as unbounded) joints, pick_ik will fail to produce solutions reliably.
Reason: get_random_valid_configuration() will sample states that have inf for joint values, since bounds are set to the numeric limits which the distribution sampler doesn't support
Behavior: FK will silently add nan, inf, or 0 to transforms which in turn fail to compute any useful distance metrics.
We should either check for these cases and warn, or implement a random sampler that supports infinite (or numeric limits) bounds.
The text was updated successfully, but these errors were encountered:
What do you think of continuous joints being handled as sampled from 0 to 2*pi radians, and once a solution is achieved snapping it to the closest multiple of 2*pi to its initial seed?
EDIT: The above suggestion assumes revolute joints, which is not necessarily the case.
When running pick_ik with robot models that have continuous (as unbounded) joints, pick_ik will fail to produce solutions reliably.
Reason: get_random_valid_configuration() will sample states that have
inf
for joint values, since bounds are set to the numeric limits which the distribution sampler doesn't supportBehavior: FK will silently add
nan
,inf
, or 0 to transforms which in turn fail to compute any useful distance metrics.We should either check for these cases and warn, or implement a random sampler that supports infinite (or numeric limits) bounds.
The text was updated successfully, but these errors were encountered: