LOONG: Online Time-Optimal Autonomous Flight for MAVs in Cluttered Environments

TL;DR

This work tackles time-critical autonomous MAV navigation in unknown clutter by introducing LOONG, a learning-accelerated online time-optimal integrated planning and control framework. LOONG combines a learning-based time allocation for time-optimal polynomial trajectory generation with a time-optimal model predictive contouring control (MPCC) that uses safe flight corridors to enforce obstacle avoidance at a variable horizon, enabling aggressive yet safe flight at 100 Hz replanning. Key contributions include a front-end A*–SFC path planner with convex polytope decomposition via CIRI, a differential-flatness-based time-optimal trajectory generator with a two-layer solve, imitation-learning for real-time time allocation, and an MPCC that tightly couples planning with full dynamics, using trajectory reuse to maintain consistency across replans. Real-world demonstrations show a peak speed of 18 m/s in cluttered environments and 10 consecutive successful trials, supported by lightweight onboard computation and LiDAR-based perception, highlighting LOONG’s practical impact for fast, safe autonomous flight in unknown environments.

Abstract

Autonomous flight of micro air vehicles (MAVs) in unknown, cluttered environments remains challenging for time-critical missions due to conservative maneuvering strategies. This article presents an integrated planning and control framework for high-speed, time-optimal autonomous flight of MAVs in cluttered environments. In each replanning cycle (100 Hz), a time-optimal trajectory under polynomial presentation is generated as a reference, with the time-allocation process accelerated by imitation learning. Subsequently, a time-optimal model predictive contouring control (MPCC) incorporates safe flight corridor (SFC) constraints at variable horizon steps to enable aggressive yet safe maneuvering, while fully exploiting the MAV's dynamics. We validate the proposed framework extensively on a custom-built LiDAR-based MAV platform. Simulation results demonstrate superior aggressiveness compared to the state of the art, while real-world experiments achieve a peak speed of 18 m/s in a cluttered environment and succeed in 10 consecutive trials from diverse start points. The video is available at the following link: https://youtu.be/vexXXhv99oQ.

Figures (11)

• Figure 1: (a) High-speed autonomous flight trajectory of our method in the real world and point-cloud map visualization. (b) The corresponding throttle and velocity profiles. The MAV completes a 20 m flight in 2.1 s and reaches a peak speed of 18.1 m/s in 1.2 s. Side and forward views both reveal that the MAV attains an aggressive pitch angle of approximately $75^\circ$ during acceleration and deceleration.
• Figure 3: Path planning and waypoints generation process involves (i) an A*-based path search, (ii) shorten and line seed generation, (iii) SFC convex decomposition (iv) and computing overlapping centers as waypoints for (v) subsequent trajectory generation.
• Figure 4: Neural network inferences time allocation from current states and waypoints to efficiently generate a multi-piece polynomial as reference trajectory.
• Figure 5: 3D MPCC diagram with local SFC constraints applied to the first $G$ steps to enable time-optimal flight while avoiding obstacles.
• Figure 6: Illustration of the strategy for maintaining consistency across two consecutive planning loops. (a) The A* algorithm is initialized from $p_G$, the $G$-th state of the trajectory generated in the previous planning loop. (b) The line segment connecting the current state $p_c$ and $p_G$ is used as the seed for generating the SFC.