Python / ROS 2 Actions

PolicyBridge-ROS2

Bounded policy execution with synchronized observations, cancellation, late-result isolation and reliable finalization.

Public source · Merged into main

Work and recorded results

• Built a ROS 2 Action runtime with synchronized RGB/joint observations, bounded backend admission, cancellation and timeout handling, and isolation of late inference results.
• Repaired diagnostic-failure finalization and snapshot expiry; 206 CPU tests passed. Validation of this repair in ROS 2 and on physical robots remains pending.

Scope of the evidence

• The earlier 204-test Humble record and four demo episodes used scripted policies and a mock manipulator; they are not physical-robot acceptance.
• Three ROS launch modules were skipped in the current Windows run. CPU doubles cover production method bodies but not ROS executor behavior.

Source and evidence

Evidence source revision: a0bcad31e358be63652ebafa36690907ae6262ab

Public source at this revision ↗

View merged repair ↗

Merge commit: bd321a2f7b216761d33b28897da49aad7794a9b5.

• Repair scope and reproducible tests
• Observation runtime

Updated 2026-09-14. CV wording and this page use the same project manifest.