HEPP: Hyper-efficient Perception and Planning for High-speed Obstacle Avoidance of UAVs
Minghao Lu, Xiyu Fan, Bowen Xu, Zexuan Yan, Rui Peng, Han Chen, Lixian Zhang, Peng Lu · 2025
Read HEPP: Hyper-efficient Perception and Planning for High-speed Obstacle Avoidance of UAVs by Minghao Lu, Xiyu Fan, Bowen Xu, Zexuan Yan, Rui Peng, Han Chen, Lixian Zhang, Peng Lu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-17438.txt.
Introduction
Minghao Lu and co-authors present HEPP, a fast perception and planning system for racing drones in clutter in a 2025 paper. Obstacle avoidance at high speed is hard: most systems fly fast only in open space or slowly in clutter. Milliseconds of delay become meters of travel.
Three modules share the work: a robot-centered mapping method far quicker than prior ones, a path search that proposes several topologically distinct routes, and a trajectory generator that pre-allocates time adaptively. The whole loop runs in milliseconds per pass at high speed in cluttered tests. Search-and-rescue and inspection drones both need this mix of speed and caution.
Catalog record
| CREATOR | Minghao Lu, Xiyu Fan, Bowen Xu, Zexuan Yan, Rui Peng, Han Chen, Lixian Zhang, Peng Lu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-17438
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
HEPP: Hyper-efficient Perception and Planning for High-speed Obstacle Avoidance of UAVs Minghao Lu, Xiyu Fan, Bowen Xu, Zexuan Yan, Rui Peng, Han Chen, Lixian Zhang, Peng Lu Abstract High-speed obstacle avoidance of uncrewed aerial vehicles (UAVs) in cluttered environments is a significant challenge. Existing UAV planning and obstacle avoidance systems can only fly at moderate speeds or at high speeds over empty or sparse fields. In this article, we propose a hyper-efficient perception and planning system for the high-speed obstacle avoidance of UAVs. The system mainly consists of three modules: 1) A novel incremental robocentric mapping method with distance and gradient information, which takes 89.5% less time compared to existing methods. 2) A novel obstacle-aware topological path search method that generates multiple distinct paths. 3) An adaptive gradient-based high-speed trajectory generation method with a novel time pre-allocation algorithm. With these innovations, the system has an excellent real-time performance with only milliseconds latency in each iteration, taking 79.24% less time than existing methods at high speeds (15 m/s in cluttered environments), allowing UAVs to fly swiftly and avoid obstacles in cluttered environments. The planned trajectory of the UAV is close to the global optimum in both temporal and spatial domains. Finally, extensive validations in both simulation and real-world experiments demonstrate the effectiveness of our proposed system for high-speed navigation in cluttered environments. I Introduction With the development of robotic autonomous technology, uncrewed aerial vehicles (UAVs) have become pivotal in various applications, ranging from surveillance to search and rescue missions [ 1 ] . Ensuring the safe and efficient navigation of UAVs is paramount for their successful deployment. Existing works have shown a robust performance in flying at a moderate speed [ 2 ] . Nevertheless, high-speed obstacle avoidance in unknown environments remains challenging and potentially hazardous. In high-speed flight scenarios, UAVs face unique challenges related to the perception accuracy and response time of the system. Typically, an autonomous obstacle avoidance system has two main components: mapping and motion planning. The function of the mapping module is to model the environment using sensor data. The limited field of view (FOV) and noise of RGBD cameras make it difficult to ensure flight safety, while Lidar sensors boast high-accuracy measurement, the extensive computation of point clouds poses challenges to real-time performance. The motion planning module generates feasible actions based on perception. When the environment is extremely cluttered and complex, it is difficult to find feasible solutions at the desired high speed. Furthermore, replanning under high-speed motion must be completed in an extremely short duration. All of these issues raise the risk of crashes when UAVs fly at high speeds in cluttered environments. Fig. 1: A demonstration of high-speed navigation, the UAV flies at over 11.0 m/s across the forest. In this article, we propose HEPP : H yper- E fficient P erception and P lanning for high-speed obstacle avoidance of UAVs, which aggregates all the mentioned difficulties in perception and planning, as shown in Fig. 1 . The overview of our proposed system is shown in Fig. 2 . The system utilizes a high-accuracy Lidar and an IMU sensor as raw data input to estimate the environment model and the UAV states, and then plan the UAV motion by a front-end topological path search and a back-end trajectory generation. For each path guiding trajectory optimization, we approximate the global optimum by selecting the one with the minimum cost. The main contributions of this article are summarized as follows: 1) We propose a novel mapping method with a new data structure. A local map can be maintained and updated with very high time and memory efficiency, and distance and gradient information can be obtained directly from the map without additional computations, which is suitable for high-resolution sensors and large-scale scenarios. 2) We propose a novel obstacle-aware topological path search algorithm. The algorithm can generate multiple non-redundant paths in complex cluttered environments with very low time cost, which facilitates the approximation towards global optimality of UAV motion planning. 3) We propose an adaptive gradient-based high-speed trajectory planning method. With the adaptive trajectory initialization, the planner can find a high-quality trajectory by numerical optimization that satisfies safety and dynamic feasibility. By selecting the best trajectory from the multiple trajectories, the UAV motion is further closer to the global optimum. 4) We propose a fully autonomous UAV system (including mapping, front-end path search, and trajectory planning) for high-speed navigation. We validate the performance of our entire system in various simulation and experimental tests. The latency of the entire system is within a few milliseconds, which is highly efficient. Fig. 2: An overview of our system framework. II Related Works II-A Local Mapping and ESDF In most robotic applications, the occupancy grid map is used for navigation and obstacle collision checks due to its intuitive principle and simple implementation [ 3 ] . The process usually consists of raycasting, updating the grid state, and so on. Based on this, the Euclidean Signed Distance Field (ESDF) [ 4 ] is widely used in gradient-based robotic trajectory optimization, where the distance and gradient information to obstacles are necessary, such as [ 5 , 6 , 7 ] . The generation of ESDF is based on the map’s data structure. In [ 8 , 9 , 10 , 11 ] , the environment is formulated by a fixed-size 3D map divided by uniform grids. Firstly, raycasting is used to update the probability value of the grids, where the state of each grid will be classified as occupied, free, or unknown. Secondly, obstacle inflation is applied to the occupied space to guarantee safety. Finally, ESDF is calculated by Breadth First Search (BFS) and trilinear interpolation. However, these uniform grid-based methods have extensive memory consumption, making them impractical for long-distance flight. Moreover, modules such as obstacle inflation, especially ESDF, will bring a very high computational load and cannot run in real-time on high-resolution Lidar sensors. Robocentric mapping methods [ 12 , 13 ] only maintain the robocentric local map using a zero-copy map sliding strategy, significantly reducing memory usage. However, these methods still have the limitation of updating ESDF in real-time. Compared to uniform grid-based data structure, octree-based mapping methods [ 14 , 15 , 16 ] are more memory efficient. The calculation of the ESDF relies on the Nearest Neighbor Search (NNS) of octree, which is also more convenient and efficient to use. However, these methods also use the fixed map origin, which will cause extensive memory usage, and re-building octree each frame is not possible in real-time. [ 17 ] proposes an incremental octree data structure for the point cloud, which is time and memory-efficient for Lidar point cloud mapping. However, this work does not present a complete methodology of mapping. Inspired by the above methods, we will propose an incremental robocentric point cloud mapping method based on [ 17 ] to address all the problems mentioned. II-B Topological Path Searching Topological path searching aims to find paths belonging to different homology classes [ 18 , 19 , 20 ] . A graph-based representation called roadmap is used to approximate the continuous configuration space. Probabilistic Roadmap (PRM) [ 21 ] is the most widely used roadmap type. [ 22 , 5 ,