From Distances to Trajectories: Real-Time Signed Distance Function Mapping and Distance-Accelerated Motion Planning for UAVs

Imagine you're building a drone that needs to navigate through a cluttered indoor space, like a warehouse filled with boxes. The challenge is to create a map of the environment while also planning a safe flight path in real time. Traditional methods often treat mapping and planning as separate tasks, which can lead to inefficiencies and potential collisions. For instance, they might use a simple occupancy grid to check for obstacles, but this can miss nuanced information about the environment, leading to unsafe trajectories. This is what's called a lack of integration between mapping and planning. To address this, the authors propose a unified approach that uses a signed distance function (SDF) to represent the environment. An SDF provides detailed information about the distance to the nearest obstacles, which is crucial for planning safe paths. They introduce an Octree Residual Network (OREN) that efficiently reconstructs SDFs from point cloud data, combining the strengths of volumetric methods and neural networks. Alongside this, they develop a planner called Bubble$^ ext{star}$ that uses the SDF to create 'bubbles' or safe zones for navigation, significantly reducing the number of collision checks needed compared to traditional methods like A$^ ext{star}$. The results show that this integrated approach allows the drone to navigate complex environments much faster and more safely than previous methods, with OREN improving SDF estimation by 22% and Bubble$^ ext{star}$ finding paths in 1-3 seconds instead of up to 10 seconds. For anyone building autonomous systems, this means a more efficient and reliable way to navigate dynamic environments.

The integration of mapping and planning using a signed distance function is a significant advancement in real-time autonomous navigation.

The approach is validated with real-time experiments on a quadrotor, showing clear improvements over existing methods.