Organizations: PRISMA Lab, Department of Engineering and Information Technology, University of Naples Federico II, Naples, Italy. · KRUSH Labs B.V., Eindhoven, Netherlands.
Visual-inertial odometry (VIO) can degrade during aerial inspection in repetitive or poorly textured environments. We propose an active environment-modification framework for unmanned aerial vehicles (UAVs) relying on cameras and an inertial measurement unit only, without prior maps or additional localization sensors. Inspired by the fairy tale of Tom Thumb, the UAV creates visual features on demand by releasing ground markers or marking walls through physical contact. An information-based indicator triggers interventions using temporal confirmation and hysteresis. Evaluation includes three simulated environments and a 900-run factorial campaign. Lower cruise speed and active marking are associated with reduced divergence, while accuracy among non-divergent runs depends on speed-strategy interactions. Corridor experiments, with five runs per configuration, use OptiTrack for flight control while VIO-based marking decisions remain active online. The combined framework reduces the reported position root mean square error by approximately 35% relative to baseline VIO, supporting active feature creation for drift mitigation.
Figures & tables
Fig. 1: Conceptual illustration of VIO drift mitigation through on-demand visual feature creation. Left: repetitive structures and insufficient visual texture can degrade VIO and increase localization drift. Centre: the UAV introduces additional visual features by releasing objects onto the ground or marking accessible walls. Right: the added features provide new visual constraints, supporting recovery of VIO estimation quality and limiting further drift accumulation.
Fig. 2: Overview of the proposed VIO-based framework. Onboard sensors feed the VIO algorithm, whose quality is continuously monitored to trigger active marking through either ground pebble deposition or wall drawing. The VIO sparse visual information is then used to build a map, on which the navigation framework in [ 23 ] generates trajectories continuously tracked by the autopilot, closing the control loop with the sensing units.
Fig. 3: Simulated scenarios: a long corridor/tunnel, a sewer shaft, and an open indoor space with sparse obstacles.
Configuration
Speed
RMSE
Max
Final
Baseline VIO
0.5 m/s
0.701
1.277
0.977
1.0 m/s
0.550
1.780
1.780
3.0 m/s
2.040
10.953
10.953
ODFC
Wall Drawing
Anticipatory
0.5 m/s
0.397
0.472
0.421
1.0 m/s
0.450
0.521
0.483
3.0 m/s
0.476
0.569
0.558
TABLE I: Sewer scenario position errors [m].
Configuration
Speed
RMSE
Max
Final
Baseline VIO
0.5 m/s
0.890
4.623
4.623
1.0 m/s
0.758
4.552
4.552
3.0 m/s
17.719
136.338
136.338
ODFC
Ground Depos.
Anticipatory
0.5 m/s
0.106
0.218
0.135
1.0 m/s
0.208
0.418
0.404
3.0 m/s
0.173
0.385
0.095
TABLE II: Open space scenario position errors [m].
Configuration
Speed
RMSE
Max
Final
Baseline VIO
0.5 m/s
0.557
1.266
0.766
1.0 m/s
0.620
1.365
0.926
3.0 m/s
0.878
1.965
1.434
ODFC – Ground Deposition
0.5 m/s
0.208
0.484
0.254
1.0 m/s
0.236
0.512
0.159
3.0 m/s
0.398
1.303
0.848
TABLE III: Corridor scenario position errors [m].
Fig. 4: (Top) Experimental setup replicating the corridor scenario in Sec. IV-A3 . (Bottom) The UAV is equipped with two sponge end-effectors for wall drawing and a bottom-mounted, PWM-controlled box for releasing pebbles on demand.
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.
Moji Shi, Rajitha de Silva, Hang Yu +2
MAVLab, TU Delft, Netherlands · Lincoln Center for Autonomous Systems (L-CAS), School of Engineering and Physical Sciences, University of Lincoln, UK
Unmanned aerial vehicles (UAVs) are becoming a powerful tool for many environmental monitoring and transport applications. Yet, their reliance on Global Navigation Satellite System (GNSS) technology for navigation makes them susceptible to catastrophic failures in scenarios where the positioning signal is unavailable or disrupted. This work explores Visual Odometry (VO) as a crucial navigation component. Recently, numerous deep-learning-based methods for image matching have been proposed that are yet to be implemented in a fully-fledged VO system. In this paper, we evaluate recent state-of-the-art image matching methods for the task of VO for UAV position tracking, with a downwards-facing camera, on our synthetic dataset, and find that while the best results are generated by the recent RoMa matcher, SIFT features can outperform some recent state-of-the-art.
Gašper Spagnolo, Luka Čehovin Zajc, Matej Dobrevski
Faculty of Computer and Information Science, University of Ljubljana, Večna pot 113, Ljubljana, Slovenia.
Visual-inertial odometry (VIO) for aerial robots relies on high rate inertial measurement unit (IMU) propagation between visual updates. However, conventional multi state constraint Kalman filters (MSCKFs) use random walk bias assumptions and fixed noise parameters, which can limit robustness when visual information is unreliable. To address this problem, we propose LBDU-VIO, a learning-augmented MSCKF with learned continuous time bias dynamics and an IMU uncertainty model. A neural ordinary differential equation (ODE) models continuous time bias dynamics to propagate the filter's bias states, replacing their random walk model. The IMU uncertainty model predicts motion adaptive measurement noise covariances for covariance propagation. Both models are trained with pose supervision without direct labels. Experiments on real world EuRoC and TUM-VI benchmarks show lower errors than representative visual-inertial baselines, including a 25.1% reduction in mean relative position error compared with S-MSCKF on EuRoC sequences with 10s visual outage.
Qizhi Guo, Junning Lyu, Defu Lin +1
School of Aerospace Engineering and the Beijing Key Laboratory of UAV Autonomous Control, Beijing Institute of Technology, Beijing 100081, China