From wildfire search and rescue to post-disaster surveys, time often means life. The mission, where every second counts, requires the UAV to fly fast and steadily in an unfamiliar and narrow environment. Professor Zhang Fu's team from the Department of Mechanical Engineering of the University of Hong Kong published a paper "Safety-assured High-speed Navigation for MAVs" in Science Robotics (2025) and proposed Safe high-speed navigation system for micro UAV-SUPER, realized 20 m/s Fly autonomously and ensure a very high success rate of obstacle avoidance, even successfully bypassing a diameter of only 2.5mm of wires, put "bird grade"The dodge instinct is brought to reality.

Video source:https://www.youtube.com/watch\?v=GPHuzG0ANmI_

01 Research background

exist unknown environment To achieve "fast and safe" flight in China, current mainstream solutions still have limitations: some racing methods mostly rely on motion capture or pre-mapping and are difficult to implement; conservative planning algorithms sacrifice speed for safety; pure visual lightweight systems are limited by range, lighting and motion blur, making it difficult to maintain stable perception at high speeds.

face challenges

• Body agility constraints High-speed obstacle avoidance requires small size and high thrust. However, after the micro aircraft is reduced in size, there is limited space left for sensors, computing units and batteries. A slight increase in load may weaken maneuverability.

• Precise sensing at long distances The effective distance of the vision/ToF solution is only a few meters, and it is difficult to detect small obstacles tens of meters away. Three-dimensional perception needs to take into account "lightweight + long range + high precision".

• Speed-Safety Contradiction The faster you fly, the easier it is to fall into a perceptual blind spot, and a switchable safe trajectory must be provided without sacrificing speed.

• Real-time performance of onboard computing power All sensing-mapping-planning links need to maintain sub-module millisecond latency within the embedded platform.

02 Technical Highlights

Focusing on the core difficulties mentioned above, this study proposed Integrated system with high mobility, long-range perception, rapid response and safety assurance-SUPER。

Image source: Ren et al., "Science Robotics" (2025), paper "Safety-assured High-speed Navigation for MAVs".

Highly agile UAV platform

The research team designed a compact quadcopter UAV with a wheelbase 280mm, take-off weight 1.5KG, thrust-to-weight ratio exceeds 5.0, with the ability to quickly turn and maneuver, meeting the high agility required for high-speed flight.

Image source: Ren et al., "Science Robotics" (2025), paper "Safety-assured High-speed Navigation for MAVs".

Precise sensing at long distances

SUPER is equipped with Livox MID360 (three-dimensional LiDAR), which has a 70-meter ranging capability and centimeter-level accuracy. It can work stably in low-light environments such as forests or at night, and is equipped with FAST-LIO2 Algorithm, integrates IMU and LiDAR data to achieve high-precision, low-latency autonomous localization and perception.

Point cloud map optimization

SUPER introduces a spatiotemporal point cloud map based on a sliding window to represent the environment state:

• Known and unknown spaces can be distinguished directly on point clouds, eliminating the ray projection overhead of traditional OGM/ESDF;

• Use the timestamp mechanism to identify dynamic obstacles and eliminate them;

• The map update delay is only 1-5ms, adapting to high-speed re-planning needs;

• It can effectively preserve the point cloud features of small objects.

Dual trajectory planning framework

SUPER uses a dual-trajectory planning framework to simultaneously calculate two trajectories in each re-planning cycle:

• Exploration trajectory: cover known and unknown areas at the same time, and increase flight speed as much as possible;

• Backup trajectory: Strictly located within a known safe area to ensure safe flight even if planning fails.

• Through the MINCO differentiable polynomial model, the trajectory shape, time allocation and switching timing are jointly optimized, and the end-to-end optimization time is controlled at 10–47 ms.

• Compared with methods such as Bubble, Raptor and Faster, SUPER's mission failure rate in simulation is reduced by 35.9-95.8 times, the average flight speed is higher, and the success rate is as high as 99.63\%.

Image source: Ren et al., "Science Robotics" (2025), paper "Safety-assured High-speed Navigation for MAVs".

03 Experimental testing

forest flight test

Verify whether SUPER can achieve high-speed, fully autonomous, and safe flight in an unknown environment without external sensory assistance. The research team is at 280×90㎡ forest area Eight sets of tests were conducted, covering lighting conditions such as day, dusk, and night.

• SUPER completed all flight missions in all tests, with a success rate of 100%, and no collision or loss of control occurred;

• When the maximum speed is set to 20m/s, stable flight is still achieved, and the average localization error is only 0.13m.

Image source: Ren et al., "Science Robotics" (2025), paper "Safety-assured High-speed Navigation for MAVs".

Comparative experiment of target tracking in complex dense forest

In order to verify whether SUPER can maintain continuous autonomous navigation and target tracking capabilities in high-density obstacle environments. Experimenters wore reflective vests and walked through two sections of woods with different densities. In some areas, they had to lower their heads and bend down to simulate a complex and dense environment. SUPER conducts comparative tests with commercial UAVs based on visual navigation to evaluate its continuous tracking and autonomous navigation capabilities of targets in woodland.

• The commercial UAV interrupted tracking twice and automatically exited the automatic mode after entering the dense forest;

• SUPER relies on point cloud maps and laser recognition functions to complete continuous tracking of running people throughout the entire process, showing stronger environmental adaptability and path planning capabilities.

Image source: Ren et al., "Science Robotics" (2025), paper "Safety-assured High-speed Navigation for MAVs".

Small obstacle obstacle avoidance experiment

In order to verify whether the system can accurately detect and avoid small targets, especially difficult-to-identify wire-type obstacles. Researchers set up wire-type obstacles with diameters ranging from 30mm to 2.5mm to verify SUPER's perception and obstacle avoidance capabilities.

• SUPER successfully identifies and avoids all obstacles, including the thinnest 2.5mm wire;

• The commercial UAV only avoided 30mm wires, and the rest collided, indicating the disadvantage of visual depth estimation in the perception of fine targets.

Image source: Ren et al., "Science Robotics" (2025), paper "Safety-assured High-speed Navigation for MAVs".

04 future outlook

In the future, SUPER is expected to continue to evolve in the directions of sensor lightweighting, system parallel optimization, and dynamic obstacle handling to further enhance its environmental adaptability. Smaller LiDAR and optimized aerodynamic design will enhance the platform's maneuverability and stability; combined with dynamic target recognition and trajectory prediction algorithms, it is expected to improve its safety in complex dynamic environments. With the characteristics of "high speed + safety", SUPER shows its outstanding performance in Post-disaster search and rescue, inspection and low-altitude transportation The application potential in such tasks is moving towards an all-weather, all-scenario intelligent flight system.

Resource Express
Paper link: https://www.science.org/doi/10.1126/scirobotics.ado6187_
open-source code: https://github.com/hku-mars/SUPER__
The content of this article is only for academic exchange and technology sharing. The copyright of the graphic and text materials belongs to the original author and the journal. If there is any infringement, please contact us to delete it. _