Vision-aided UAV navigation and dynamic obstacle avoidance using gradient-based B-spline trajectory optimization

TL;DR

This paper tackles real-time UAV navigation in dynamic environments by fusing vision-based 3D dynamic mapping with gradient-based B-spline trajectory optimization. The ViGO framework introduces a circle-based guide-point method to rapidly estimate static collision costs and a receding-horizon distance field to account for moving obstacles, all wrapped in an iterative re-guide loop to guarantee collision-free trajectories. Key contributions include the vision-aided 3D dynamic map, the efficient static-cost estimator, and the RH-based dynamic-cost formulation, which together yield robust, real-time performance on onboard hardware and higher success rates than state-of-the-art planners. The practical impact lies in enabling lightweight UAVs to safely navigate crowded, changing environments without relying on pre-built maps, with open-source software to foster adoption and further development.

Abstract

Navigating dynamic environments requires the robot to generate collision-free trajectories and actively avoid moving obstacles. Most previous works designed path planning algorithms based on one single map representation, such as the geometric, occupancy, or ESDF map. Although they have shown success in static environments, due to the limitation of map representation, those methods cannot reliably handle static and dynamic obstacles simultaneously. To address the problem, this paper proposes a gradient-based B-spline trajectory optimization algorithm utilizing the robot's onboard vision. The depth vision enables the robot to track and represent dynamic objects geometrically based on the voxel map. The proposed optimization first adopts the circle-based guide-point algorithm to approximate the costs and gradients for avoiding static obstacles. Then, with the vision-detected moving objects, our receding-horizon distance field is simultaneously used to prevent dynamic collisions. Finally, the iterative re-guide strategy is applied to generate the collision-free trajectory. The simulation and physical experiments prove that our method can run in real-time to navigate dynamic environments safely. Our software is available on GitHub as an open-source package.

Figures (9)

• Figure 1: The UAV navigating with obstacles using the proposed algorithm. The upper presents the physical autonomous flight in a dynamic environment. The bottom shows our customized UAV with onboard sensors.
• Figure 2: Illustration of the 3D dynamic map. (a) Depth image with detected objects. (b) RGB camera view. (c) U-depth map generated by the depth image with the detected results. (d) Visualization of the dynamic map.
• Figure 3: Illustration of circle-based guide-point assignment. For the given collision trajectory, we first find the collision control points and search the collision-free paths. The guide points shown as purple points are the intersections between circle-based raycasting and the searched paths.
• Figure 4: Illustration of receding horizon distance field. We linearly decrease the safety distance $\text{r}$ from the current obstacle position $\textbf{O}_{0}$ to the last predicted position $\textbf{O}_{k}$ in a receding horizon manner.
• Figure 5: The recorded average runtime for each component of our system. The entire system is able to run in real-time by the onboard computer.