OctoMap vs Voxblox: Probabilistic Occupancy vs. TSDF for Robotic Mapping

OctoMap excels at providing a probabilistic, memory-efficient 3D occupancy map, making it the standard for long-term, static environment modeling. Its octree-based structure explicitly represents free, occupied, and unknown space, which is critical for conservative path planning. For example, a drone navigating a cluttered warehouse can use OctoMap to store a persistent map with a memory footprint often 10-50x smaller than a dense point cloud, while explicitly reasoning about sensor noise and dynamic obstacles through probabilistic occupancy updates.

Voxblox takes a fundamentally different approach by building a Euclidean Signed Distance Field (ESDF) from a Truncated Signed Distance Field (TSDF). This strategy prioritizes speed for local trajectory optimization. By pre-computing the distance to the nearest obstacle for every voxel, a planner can instantly query the collision gradient. This results in a trade-off: Voxblox generates collision costs in microseconds, enabling fast Model Predictive Control (MPC), but it is less memory-efficient for mapping vast, empty spaces compared to OctoMap's sparse octree structure.

The key trade-off: If your priority is building a globally consistent, long-term map with explicit uncertainty for high-level navigation, choose OctoMap. If you prioritize microsecond-level collision checking for aggressive, real-time local trajectory planning, choose Voxblox. For a full autonomous stack, many engineers use OctoMap for the global planner and Voxblox for the local planner.

OctoMap excels at probabilistic occupancy mapping because its octree-based structure inherently models free, occupied, and unknown space with a mathematically rigorous Bayesian update scheme. For example, in dynamic environments where sensor noise or temporary occlusions are common, OctoMap's ability to represent the probability of occupancy allows a drone to plan conservative paths that avoid uncertain regions, a critical feature for safety-certified flight stacks. This makes it the standard for applications requiring explicit reasoning about unexplored space.

Voxblox takes a different approach by building Truncated Signed Distance Fields (TSDFs) and Euclidean Signed Distance Fields (ESDFs) directly from sensor data. This results in a map that is optimized for speed: the ESDF provides instantaneous gradient information, enabling gradient-based trajectory optimization for collision avoidance. In a benchmark on a ground robot platform, Voxblox demonstrated ESDF update times that scale with the surface area of new sensor data rather than the volume of the map, making it significantly faster than OctoMap for local replanning in cluttered scenes.

The key trade-off: If your priority is explicit uncertainty modeling for conservative, long-term navigation in noisy environments, choose OctoMap. Its probabilistic framework is essential for missions where knowing what you don't know is as important as knowing the obstacles. If you prioritize fast, local collision checking for agile flight or manipulation, choose Voxblox. The continuous distance field enables smoother, more efficient trajectories that are computed in real-time, but it sacrifices the explicit representation of unknown space.