Feature Request: Success-on-Any Parallel Composite

[coolbeevip]

Feature Request: Success-on-Any Parallel Composite

Summary

I would like to propose a new parallel composite behavior that executes all children concurrently and succeeds as soon as any child succeeds, regardless of whether other children fail.

Use Case

In many real-world scenarios, we need to attempt multiple approaches simultaneously and consider the task successful if any one approach succeeds. For example:

• Robot Navigation: Try multiple path-finding algorithms simultaneously, succeed if any finds a valid path
• Network Requests: Make parallel requests to multiple servers, succeed if any responds successfully
• Sensor Fusion: Query multiple sensors concurrently, succeed if any provides valid data
• Backup Systems: Run primary and backup processes in parallel, succeed if either completes successfully

Current Limitations

The existing Parallel composite with SuccessOnOne policy doesn't meet this requirement because:

• It returns FAILURE immediately if any child fails
• This prevents the "succeed if any succeeds" behavior when some children might legitimately fail

The current policies behave as follows:

• SuccessOnAll: Requires all children to succeed
• SuccessOnOne: Succeeds if one succeeds AND no others fail
• SuccessOnSelected: Requires specific children to succeed

[erichlf]

The documentation states:

Parallels with policy :class:~py_trees.common.ParallelPolicy.SuccessOnOne
return :py:data:~py_trees.common.Status.SUCCESS if at least one child
returns :py:data:~py_trees.common.Status.SUCCESS and others are
:py:data:~py_trees.common.Status.RUNNING

and the code seems to agree:

elif type(self.policy) is common.ParallelPolicy.SuccessOnOne:
                for child in reversed(self.children):
                    if child.status == common.Status.SUCCESS:
                        new_status = common.Status.SUCCESS
                        self.current_child = child
                        break

But I guess the docs seem to say it will return SUCCESS as soon as the first one returns success. The code definitely seems to return SUCCESS as soon as there is a SUCCESS and it doesn't matter what is going on with the other behaviors.

[coolbeevip]

Thank you for your discussion.

I noticed that as long as any child node returns FAILURE, the Parallel node directly returns FAILURE, and there will be no further evaluation of the SuccessOnOne/SuccessOnAll/SuccessOnSelected strategies.

https://github.com/splintered-reality/py_trees/blob/devel/py_trees/composites.py#L719-L728

[erichlf]

Yep, seems counter-intuitive.

[coolbeevip]

This might be a case of code comments being out of sync. I found the following description at https://py-trees.readthedocs.io/en/devel/composites.html, which mentions:

Parallels will return FAILURE if any child returns FAILURE.

Parallels with policy SuccessOnOne return SUCCESS if at least one child returns SUCCESS and others are RUNNING.

[coolbeevip]

Hi @sea-bass, just following up on this feature request. Is there any update or feedback you could share? If you need any additional information from my side, I'm happy to provide it. Thanks!

[sea-bass]

This is not on my radar at all, as I'm just helping maintain the tool but not really doing any core development. But I'm happy to review PRs as well!

I would also suggest a "workaround" for now -- which is to put FailureisRunning decorators on all the child nodes. Along with a SuccessOnOne policy, this should achieve what you want for now?