Voxblox: Incremental 3D Euclidean Signed Distance Fields for On-Board MAV Planning (1611.03631v2)

Abstract: Micro Aerial Vehicles (MAVs) that operate in unstructured, unexplored environments require fast and flexible local planning, which can replan when new parts of the map are explored. Trajectory optimization methods fulfill these needs, but require obstacle distance information, which can be given by Euclidean Signed Distance Fields (ESDFs). We propose a method to incrementally build ESDFs from Truncated Signed Distance Fields (TSDFs), a common implicit surface representation used in computer graphics and vision. TSDFs are fast to build and smooth out sensor noise over many observations, and are designed to produce surface meshes. Meshes allow human operators to get a better assessment of the robot's environment, and set high-level mission goals. We show that we can build TSDFs faster than Octomaps, and that it is more accurate to build ESDFs out of TSDFs than occupancy maps. Our complete system, called voxblox, will be available as open source and runs in real-time on a single CPU core. We validate our approach on-board an MAV, by using our system with a trajectory optimization local planner, entirely on-board and in real-time.

• The paper presents a novel method that fuses TSDF data with an anti-grazing heuristic to create high-fidelity 3D distance fields in real time.
• The paper leverages a queue-based propagation algorithm to convert TSDFs into ESDFs, thereby enhancing collision detection and trajectory optimization.
• The paper validates its approach through extensive experiments on varied datasets, achieving order-of-magnitude performance gains in MAV planning.

Voxblox: Building 3D Signed Distance Fields for Planning

The paper presents "Voxblox," a framework for constructing Truncated Signed Distance Fields (TSDFs) and Euclidean Signed Distance Fields (ESDFs) in real-time on robotic platforms. While TSDFs are traditionally used in high-resolution 3D reconstructions, their application for robotic planning has been limited by computational and memory demands. This research proposes strategies to alleviate these constraints by focusing on large voxel sizes that balance computation speed and environment modeling fidelity.

TSDF Integration and Performance Enhancements

The authors emphasize the need for effective TSDF implementation for use in path planning. They acknowledge the computational intensity traditionally associated with TSDFs due to small voxel sizes, previously deemed necessary for accurate real-time processing. To address this, they introduce a novel merging strategy that not only reduces the time complexity by aggregating ray intersections per voxel but also preserves geometric accuracy through a refined anti-grazing heuristic. This enhancement prioritizes surface boundaries and mitigates inaccuracies caused by grazed rays in large voxels. The improvements are substantiated through experimental validation against prior methods, revealing order-of-magnitude performance gains.

ESDF Construction from TSDF

In planning contexts, access to Euclidean distance information is indispensable for trajectory optimization and collision detection. The paper builds upon algorithms for dynamic ESDF construction and adapts them to process TSDF-originated distance data. This adaptation allows the framework to pre-compute critical spatial metrics while maintaining a flexible map size—a feature particularly advantageous for dynamic, unpredictable environments. The authors utilize queue-based propagation mechanisms to efficiently update distance fields, exploiting the inherent proximity information provided by TSDFs.

Experimental Evaluation and Practical Implications

The researchers demonstrate Voxblox's viability across multiple datasets, encompassing indoor and outdoor scenarios captured by stereo and RGB-D sensors. Performance benchmarks illustrate the framework's capability in handling diverse environments, confirming both structural accuracy and a substantial reduction in processing overhead compared to traditional methods. These results manifest practical strides toward deploying real-time onboard planning systems, suggesting broader applicability in autonomous navigation tasks where computational resources are limited.

Future Directions in AI and Robotics

Voxblox lays foundational work for real-time, resource-efficient spatial modeling in robotics. Future research directions may explore the integration of Voxblox with advanced optimization-driven planning algorithms, enhancing autonomous systems' decision-making capabilities in dynamic environments. Moreover, the principles outlined in this paper can inspire further investigation into balancing precision and efficiency in spatial representations beyond the robotics domain, impacting areas like augmented reality and computational imaging.

In conclusion, "Voxblox" presents an effective approach to integrating TSDFs and ESDFs for real-time planning, making significant contributions to the field of robotic mapping and planning. The methods developed have the potential to drive advancements in autonomous capabilities, offering a scalable solution to environmental modeling challenges inherent in robotics and AI.