Polka v0.5 boosts lidar processing with 6.2x faster deskewing
New coarse-stride SE(3) interpolation cuts deskew latency from 9.8ms to 1.6ms.
Panav announced Polka v0.5.0, a major update to their all-in-one ROS 2 node for 2D and 3D lidar processing. The headline improvement is a 6.2x speedup in deskewing: by using coarse-stride SE(3) rotation interpolation, per-source deskew latency drops from ~9.8 ms to ~1.6 ms. Angular filtering (CPU-based atan2 rejection) sees a 3x boost, cutting tick time from 10.47 ms to 3.55 ms at 259k points from three Robosense Airy lidars. The new IMU estimation feature uses EMA to infer body-frame gravity when the IMU lacks orientation data, enabling gravity-aware deskewing even on lower-cost sensors. A bug fix from @Kyrie_James for Robosense Airy also improved angular deskew accuracy.
Operationally, v0.5 introduces live parameter tuning: users can change parameters on a running node without restart, using a two-phase apply that ensures pipeline consistency during swaps. A new /diagnostics stream provides timing-drift and rate-drift flags the moment a source degrades. The Dashboard now visualizes sensor issues, current bandwidth, topic frequency, deployed config, and warnings (or can be disabled for classic logging). Polka v0.5 expands support to five ROS 2 distros (Humble, Jazzy, Iron, Kilted, Lyrical) and includes performance measurement scripts. Upgrade notes: suppress_duplicate_timestamps and diagnostics.enabled default to true, but existing configs and launch files remain compatible.
- Deskew latency reduced from 9.8ms to 1.6ms using coarse-stride SE(3) rotation interpolation.
- Angular filter 3x faster, dropping tick time from 10.47ms to 3.55ms at 259k points (3 lidars).
- New sensor dashboard with live parameter tuning, diagnostics stream, and support for 5 ROS 2 distros.
Why It Matters
Real-time multi-lidar fusion just got cheaper and faster, enabling high-performance autonomy on lower-end hardware.