Existing aerial robot navigation systems typically plan paths around static and dynamic obstacles, but fail to adapt when a static obstacle suddenly moves. Integrating environmental semantic awareness enables estimation of potential risks posed by suddenly moving obstacles. In this paper, we propose RA- Nav, a risk-aware navigation framework based on semantic segmentation. A lightweight multi-scale semantic segmentation network identifies obstacle categories in real time. These obstacles are further classified into three types: stationary, temporarily static, and dynamic. For each type, corresponding risk estimation functions are designed to enable real-time risk prediction, based on which a complete local risk map is constructed. Based on this map, the risk-informed path search algorithm is designed to guarantee planning that balances path efficiency and safety. Trajectory optimization is then applied to generate trajectories that are safe, smooth, and dynamically feasible. Comparative simulations demonstrate that RA-Nav achieves higher success rates than baselines in sudden obstacle state transition scenarios. Its effectiveness is further validated in simulations using real- world data.
Figures & tables
Fig. 1: Framework of the semantic-enhanced navigation system. The perception and planning modules run asynchronously on the onboard computer. The perception module uses U-Net for semantic segmentation and updates a local map at fixed intervals. The planning module then uses the risk grid map for safe path search and optimization.
Fig. 2: Semantic segmentation network architecture, composed of a 2D backbone and a 3D semantic segmentation head, which enables multi-scale semantic segmentation.
Fig. 3: (a) is the risk illustration of the static obstacle, its center is (0,0) and its width and length are 5. (b) is the risk illustration of the dynamic obstacle, its center is (0,0) and its width and length are 2. The risk along the velocity of the dynamic obstacle is higher than the opposite direction.
Fig. 4: Local coordinate system centered at the obstacle. When Δ>0 , the position lies in front (quadrants I and IV); when Δ<0 , the position is behind (quadrants II and III).
Fig. 5: Initialization of control points. When the risk value of an obstacle exceeds the threshold Rthresh , the control points are shifted along the direction of risk reduction.
Method
IoU
mIoU
Prec.
FPS
SCPNet [ 29 ]
56.10
36.70
72.43
< 1
VoxFromer-S [ 28 ]
57.54
16.48
70.85
< 1
SSCNet [ 30 ]
53.20
14.55
59.13
12.00
LMSCNet [ 19 ]
55.32
17.01
77.11
13.50
TABLE I: Comparison of different methods on semantic segmentation metrics
Fig. 6: Comparison of A* path search results in high-risk static environment. The random dimensionless map size is 50×50, where black indicates static obstacles, blue denotes predefined high-risk obstacles with motion capability, the green point represents the start, and the red point represents the goal.
Fig. 7: Risk potential field and corresponding risk gradient of the random map, where red arrows indicate the direction of risk gradient descent.
Scene
Method
Path Length
Minimum Safe Distance
Static Env.
R-A*(ours)
61.8
2.83
A* [ 21 ]
59.5
1
Dynamic Env.
R-A*(ours)
65.25
2.06
A* [ 21 ]
62.33
0
D* [ 22 ]
66.33
0.5
TABLE II: Path search result comparison in high-risk environment
Fig. 8: Environmental risk field and its gradient corresponding to the obstacle in motion.
Fig. 9: Comparison of path search results in high-risk dynamic environment. When a search node approaches a moving obstacle (marked in blue), the obstacle moves with a velocity of (−0.5,−0.3) .
Fig. 10: Comparison of trajectories with the baseline in the same scenario. (a) Simulation environment in Gazebo. (b) Our method generates a safe trajectory by accounting for potential obstacle risks. (c) Ego-Planner fails to assess such risks and produces a trajectory close to the obstacle with potential collision risk.
Fig. 11: Trajectory comparison with the baseline in the dynamic scenario. (a) In the Gazebo simulation, the ‘person’ starts moving when the Aerial Robot approaches, resulting in a collision. (b) Ego-Planner generates an initial trajectory close to the obstacle, and when the ’person’ moves, it fails to react in time, causing a collision. (c) In the Gazebo simulation, the Aerial Robot successfully avoids the ’person’ without collision. (d) Our planner accounts for the potential risk of the ’person’ and generates an initial trajectory away from the obstacle, enabling timely avoidance when the ’person’ moves.
Fig. 12: Trajectory comparison in the same outdoor simulation scenario. (a) In the Gazebo environment, the ‘person’ is occluded by a house and starts moving when the Aerial Robot approaches. (b) Ego-Planner lacks semantic understanding of the environment and fails to account for the occlusion risk, resulting in a trajectory close to the obstacle. (c) Our planner leverages house semantics to construct a risk field in advance and generates a safe trajectory away from the obstacle.
Fig. 13: Trajectory comparison when the Aerial Robot encounters the ‘person’. (a) Ego-Planner generates a trajectory close to the house, leading to a collision when the ‘person’ continues to move. (b) Our planner generates a trajectory away from the house, providing sufficient reaction distance for the Aerial Robot to avoid the obstacle when the ‘person’ continues to move.
Scene
Method
Success Rate
Planning Time
Flight Time
(%)
(ms)
(s)
1
Ego-planner [ 2 ]
25
0.113
3.50
Ours (only R-A*)
70
0.447
4.29
Ours (full)
100
0.727
6.09
2
Ego-planner [ 2 ]
20
0.165
6.56
Ours (only R-A*)
75
0.368
6.91
TABLE III: Comparison of success rate, planning time, and flight time in two simulated scenes.
Fig. 14: RA-Nav can predict potential risks in occluded environments based on semantic information and plan collision-free trajectories.
Safe robot navigation develops methods for reaching goals while avoiding collisions and managing uncertainty, dynamic obstacles, and environmental hazards.
The problem of autonomous navigation for UAV inspection remains challenging as it requires effectively navigating in close proximity to obstacles, while accounting for dynamic risk factors such as weather conditions, communication reliability, and battery autonomy. This paper introduces the MOAR path planner which addresses the complexities of evolving risks during missions. It offers real-time trajectory adaptation while concurrently optimizing safety, time, and energy. The planner employs a risk-aware cost function that integrates pre-computed cost maps, the new concepts of damage and insertion costs, and an adaptive speed planning framework. With that, the optimal path is searched in a graph using a discrete representation of the state and action spaces. The method is evaluated through simulations and real-world flight tests. The results show the capability to generate real-time trajectories spanning a broad range of evaluation metrics: around 90% of the range occupied by popular algorithms. The proposed framework contributes by enabling UAVs to navigate more autonomously and reliably in critical missions.
Louis Petit, Alexis Lussier Desbiens
the Createk Design Lab, University of Sherbrooke, Sherbrooke, J1K 2R1, Canada
Navigation using a monocular camera is pivotal for autonomous operation on tiny aerial robots due to their perfect balance of versatility, cost and accuracy. In this paper, we introduce MinNav, a navigation stack based on optical flow and its uncertainty to fly through a scene with static and dynamic obstacles and unknown-shaped gaps without any prior knowledge of the scene components and/or their locations/ordering. We further improve success rate by using the activeness of the robot to move around in an exploratory way to find obstacles and navigate. We successfully evaluate and demonstrate the proposed approach in many real-world experiments in various environments with static and dynamic obstacles and unknown-shaped gaps with an overall success rate of 70%. To the best of our knowledge, this is the first solution to tackle all the aforementioned navigation cases without prior knowledge using a monocular camera. Our approach is on par in performance with depth based methods with factors of magnitude less computation required and can readily run onboard tiny aerial robots. The accompanying video, supplementary material, code and dataset can be found at https://pear.wpi.edu/research/minnav.html
Obstacle avoidance for autonomous quadrotors in cluttered environments remains challenging, especially when both dense static obstacles and moving obstacles are present. Existing methods are often tailored to either static or dynamic settings, which always limit their adaptability across diverse environments. This paper presents TransNav, a transformer-based end-to-end navigation method for quadrotor obstacle avoidance in both static and dynamic environments. TransNav employs a patch-based point-cloud encoding scheme that combines local feature extraction with global contextual modeling, enabling compact yet informative environment representation. The method integrates perception and navigation in a single architecture and directly predicts a finite-horizon action sequence, from which the first action is executed. Moreover, we develop an unsupervised learning strategy by incorporating a lightweight point-mass dynamics model, which directly optimizes trajectory-level costs over the predicted sequence through numerical gradient propagation. Thus, the policy can be efficiently trained without expert demonstrations. Simulation and real-world experiments show that the proposed method achieves strong performance in dense static environments and competitive obstacle-avoidance capability in dynamic scenarios, while maintaining low inference cost for onboard deployment.
Jianghan Zhu, Zhuo Li, Yanjun Zhang +1
The authors are with the State Key Laboratory of Autonomous Intelligent Unmanned Systems, School of Automation, Beijing Institute of Technology, Beijing 100081, China