Perception-Aware Autonomous Exploration in Feature-Limited Environments
Authors: Moji Shi, Rajitha de Silva, Hang Yu, Riccardo Polvara, Marija Popović
Organizations: MAVLab, TU Delft, Netherlands · Lincoln Center for Autonomous Systems (L-CAS), School of Engineering and Physical Sciences, University of Lincoln, UK
Autonomous exploration in unknown environments typically relies on onboard state estimation for localisation and mapping. Existing exploration methods primarily maximise coverage efficiency, but often overlook that visual-inertial odometry (VIO) performance strongly depends on the availability of robust visual features. As a result, exploration policies can drive a robot into feature-sparse regions where tracking degrades, leading to odometry drift, corrupted maps, and mission failure. We propose a hierarchical perception-aware exploration framework for a stereo-equipped unmanned aerial vehicle (UAV) that explicitly couples exploration progress with feature observability. Our approach (i) associates each candidate frontier with an expected feature quality using a global feature map, and prioritises visually informative subgoals, and (ii) optimises a continuous yaw trajectory along the planned motion to maintain stable feature tracks. We evaluate our method in simulation across environments with varying texture levels and in real-world indoor experiments with largely textureless walls. Compared to baselines that ignore feature quality and/or do not optimise continuous yaw, our method maintains more reliable feature tracking, reduces odometry drift, and achieves on average 30% higher coverage before the odometry error exceeds specified thresholds.
Figures & tables
Fig. 1: Our approach for perception-aware exploration applied on a UAV. (a) Odometry trajectory during an exploration task (blue arrows); arrow headings indicate the planned yaw angle. (b) Corresponding camera view. Our method maintains reliable feature tracking during exploration.
Fig. 2: Our perception-aware exploration framework. Two key components are frontier selector and perception-aware trajectory optimisation.
Fig. 3: Our frontier selector. Frontier voxels are clustered by proximity, and candidate viewpoints are sampled from known free space to observe the frontiers. Visual features are overlaid, where blue intensity indicates feature quality: textured obstacles yield more and higher-quality features than textureless surfaces. Our selector prefers bottom-left viewpoints, which are farther from the current UAV position but provide stronger feature support than closer viewpoints.
Fig. 4: Our yaw angle trajectory optimisation procedure. Dashed field of view sectors represent the waypoint yaws {ψj} and arrows with different colours show covisible feature sets {Fj} . The zoomed-in view shows the calculation of the relative angular velocity. For any point on trajectory, we decompose the velocity pj˙(t) in the direction of the feature vectors fi−pj(t) and the perpendicular direction, expecting the yaw rate to compensate the relative angular velocity.
Fig. 6: Frame-averaged spatial distribution of tracked visual features in the image plane during exploration. The axes show the pixel coordinates (u,v) . For each frame, we build a 2D histogram of tracked feature locations and then average the histogram over time across the entire exploration run; colour indicates the normalised mean feature count per bin.
Fig. 7: Number of tracked visual features per frame during exploration. The x -axis shows normalised exploration progress, defined as the explored free-space volume divided by the total explored volume at run completion. The y -axis reports the number of features successfully tracked per frame. Solid lines show the mean over 10 runs; shaded areas indicate one standard deviation.
Method
High-Texture Map Exploration Rate
Medium-Texture Map Exploration Rate
Low-Texture Map Exploration Rate
1.0 m
2.0 m
3.0 m
1.0 m
2.0 m
3.0 m
1.0 m
2.0 m
3.0 m
Ours
0.913 ± 0.065
0.913 ± 0.065
0.924 ± 0.049
0.888 ± 0.070
0.928 ± 0.044
0.928 ± 0.044
0.326 ± 0.169
0.427 ± 0.126
0.470 ± 0.139
FUEL
0.728 ± 0.291
0.846 ± 0.207
0.868 ± 0.182
0.651 ± 0.311
0.731 ± 0.257
0.802 ± 0.175
0.224 ± 0.078
0.271 ± 0.100
0.317 ± 0.128
FUEL + PA Frontier
0.815 ± 0.085
0.828 ± 0.058
0.828 ± 0.058
0.667 ± 0.147
0.771 ± 0.093
0.771 ± 0.093
0.321 ± 0.202
0.337 ± 0.206
0.325 ± 0.203
LA
0.234 ± 0.094
0.235 ± 0.093
0.235 ± 0.093
0.202 ± 0.093
0.220 ± 0.082
0.239 ± 0.093
0.142 ± 0.064
0.162 ± 0.071
0.176 ± 0.087
LA + PA Frontier
0.023 ± 0.005
0.023 ± 0.005
0.023 ± 0.005
0.048 ± 0.020
0.048 ± 0.020
0.048 ± 0.020
0.052 ± 0.033
0.053 ± 0.034
0.054 ± 0.035
TABLE I: Success exploration rate (expressed as mean ± std dev) under different odometry error thresholds ( 1.0m , 2.0m , 3.0m ) on environments ordered by texture richness (High → Low). Best per column in bold .
Fig. 8: Exploration process in Medium-Texture map. Blue and green lines indicate the odometry and ground-truth pose. Blue arrows indicate yaw angles. Our perception-aware approach enables looking at known features (bottom-left).
Fig. 9: Feature tracking comparison for ablation study. Brighter colours indicate a higher number of tracked features in the image plane. Removing either key component reduces feature intensity, indicating degraded feature tracking.
Method
Low-Texture Map Exploration Rate
1.0 m
2.0 m
3.0 m
Ours
0.326 ± 0.169
0.427 ± 0.126
0.470 ± 0.139
Ours w/o PA Frontier
0.162 ± 0.123
0.185 ± 0.131
0.209 ± 0.130
Ours w/o Yaw Optim.
0.300 ± 0.057
0.368 ± 0.047
0.404 ± 0.059
TABLE II: Exploration success rate comparison for ablation study under different odometry error thresholds. Best in bold .
Map
Mean total feature number
Mean error (m)
Sparse Layout
519.7
1.0188
Dense Layout
969.3
0.8204
TABLE III: Results across two real-world layouts.
Fig. 10: Real-world experiment results in Dense Layout . (a) Part of the exploration process from top-down view. White arrows show the camera directions. (b) Camera field of view. Orange denotes visual features currently tracked by the UAV, and white denotes other obstacles. (c) Global feature map overlaid on the occupancy map: feature quality is colour-coded (blue: low score, red: high score), illustrating that high-quality features concentrate on textured obstacles.
Autonomous exploration in unknown environments requires unmanned aerial vehicles (UAVs) to efficiently generate informative trajectories while simultaneously constructing accurate maps. Although many existing exploration methods rely on geometric information, they often lack semantic awareness, resulting in suboptimal exploration efficiency and limited environmental understanding. To address this limitation, this paper proposes a semantic-aware exploration framework that adds semantic information to a roadmap-based exploration strategy. The proposed method builds on the Dynamic Exploration Planner (DEP), which incrementally constructs a Probabilistic Roadmap (PRM), and augments this roadmap with a semantic layer. A semantic reward function is introduced to prioritize regions containing meaningful objects and structures, enabling the UAV to focus on areas with higher information value. Furthermore, the roadmap is continuously updated to support efficient frontier selection and path planning during exploration. The proposed framework is implemented in ROS Noetic and Gazebo using an RGB-D sensor for simultaneous acquisition of geometric and semantic information. Experimental results in multiple simulated environments demonstrate that the proposed approach achieves exploration coverage rates between 90% and 94% while reducing exploration time and travel distance compared with conventional geometry-based exploration methods.
Duc-Thien Nguyen, Ngoc Minh Do, Xiem HoangVan +1
University of Engineering and Technology, Vietnam National University, 10000, Hanoi, Vietnam. · School of Information Science, Japan Advanced Institute of Science and Technology, Nomi, 923-1211, Ishikawa, Japan.
Visual navigation in unknown environments remains a core challenge in mobile robotics, especially for resource-constrained platforms. Most existing approaches rely on loosely coupled modular pipelines and strong assumptions on perception quality or environmental structure, often resorting to multi-modal sensor suites that increase system complexity and deployment cost. Vision-only navigation offers a lightweight alternative, but its performance degrades severely under motion blur, low texture, and illumination changes, largely because they neglect the tight coupling between commanded motion and perception. While perception-aware methods partially address this issue, they typically optimize individual modules and fail to propagate uncertainty consistently across the navigation stack. In this paper, we present UNSEEN, a unified uncertainty- and perception-aware navigation framework that explicitly couples localization, mapping, and planning using only a front-mounted camera. UNSEEN estimates sparse maps and robot poses with associated uncertainties at 6Hz, and leverages them to plan trajectories that jointly optimize task progress and estimation accuracy in receding-horizon. Simulations and extensive real-world experiments in unknown environments demonstrate the robustness of the proposed approach, with UNSEEN-SLAM reducing absolute translational error by 9.8% and UNSEEN-Plan improving estimation accuracy by up to 45% compared to state-of-the-art methods, while achieving a 100% task success rate.
Tommaso Faraci, Marco Camurri, Daniele Fontanelli +1
Classic exploration methods often rely on dense occupancy maps or high-resolution point clouds for frontier detection and path planning, resulting in substantial memory consumption and computational overhead. Moreover, micro UAVs under size, weight, and power (SWaP) constraints are not practical to be equipped with sensors like LiDAR to obtain accurate environmental geometric measurements. This paper presents a lightweight autonomous exploration system that leverages omnidirectional vision and sparse topological map guidance. Specifically, we utilize a multi-fisheye camera setup to achieve omnidirectional Field of View (FoV) and perform depth estimation. To address the limited depth estimation accuracy, frontiers are represented as potential unexplored regions characterized by topological nodes instead of explicit boundaries, enabling efficient identification of frontier regions without maintaining occupancy grids or global point clouds. Unlike classic dense representations, our approach abstracts the environment using a sparse topological map composed of key nodes and their descriptors, reducing memory consumption and computational demands. Global path planning is performed directly on the sparse graph. The proposed method is validated in both simulation and on a palm-sized vision-based UAV with an 11 cm wheelbase and a 400 g weight in real-world experiments, demonstrating that our method can achieve efficient exploration with extremely low computational consumption.
Zirui Wang, Xinjia Luo, Haotian Sun +3
Department of Mechanical and Energy Engineering, Southern University of Science and Technology, Shenzhen, 518055, China