Air–Ground Cooperative Precision Landing

• Air–ground cooperative precision landing is a method where UAVs and ground agents work together using sensor fusion and dynamic planning to achieve centimeter-level landing precision.
• Advanced system architectures integrate visual, IR, and minimal sensor configurations with control strategies like FSM, MPC, and photodiode-based servoing to optimize landing performance.
• Experimental studies demonstrate sub-decimeter accuracy and robust performance by leveraging real-time joint optimization and bi-directional control in complex, dynamic environments.

Air–ground cooperative precision landing encompasses techniques and algorithms in which an aerial vehicle (typically a UAV or drone) and a ground agent (static or mobile) act in coordinated roles to achieve high-accuracy autonomous landing, often under resource or environmental constraints. In this paradigm, both agents—rather than being treated as independent or, in the canonical case, passive/active—may cooperate by sharing perception, dynamically planning trajectories, or adapting platform states (e.g., mobile platform motion or deck orientation) to facilitate robust, safe, and efficient recovery. Air–ground cooperation enables sub-meter to centimeter-scale accuracy for applications ranging from mobile recharging to dynamic recovery in challenging scenarios such as high winds, moving targets, or denied GNSS environments.

1. System Architectures and Sensing Modalities

State-of-the-art cooperative landing systems integrate advanced sensing platforms on both UAV and ground agent, ranging from monocular vision and IR arrays to gimbal-stabilized multimodal payloads and microcontroller-driven photodiode assemblies.

Key architectural elements:

• Multi-Sensor Gimbal Payloads: The approach in "A Precision Drone Landing System using Visual and IR Fiducial Markers and a Multi-Payload Camera" (Springer et al., 2024) realizes air–ground cooperation via active sensor switching between zoom, wide-angle, and IR modalities. The drone's gimbal executes independent pan/tilt commands, dynamically aligning sensors with ground fiducials under a finite-state machine policy.
• Minimal Sensor Microdrones: AIRA ("A Low-cost IR-based Approach...") pioneers sub-10g low-power landing for resource-constrained drones by equipping three photodiodes and a servo; the ground station emits a strong IR field detected by vector summation on the drone (Liu et al., 2024). This configuration achieves landing precision below 10 cm over 11 m with total sensing power under 0.1 mW.
• Pose and Kinematic Coordination: High-dexterity drones leverage onboard monocular cameras (as in (Salagame et al., 2022)) with visual fiducials to estimate 6-DoF relative pose, while ground platforms may be mobile and capable of reacting to aerial cues. Systems incorporate state estimation pipelines fusing vision, IMU, and platform odometry, and may include explicit platform attitude control (deck tilt) for parallelized alignment (Zhao et al., 6 Jan 2026).

2. Cooperative Control and Trajectory Planning

Coordination protocols span simple PID-based tracking to bi-directional, nonlinear, model-predictive frameworks that jointly optimize air and ground trajectories in real time.

• Finite-State Machine and Bearing-Only Servoing: The system in (Springer et al., 2024) exploits a state-machine policy operating over search, tracking, approach, and recovery modes. Guidance is based on minimal bearing (pan/tilt/yaw) and pixel-size cues, eliminating the need for explicit range or altitude estimation and permitting robust operation—even during temporary pad occlusion.
• Complementarity-Constrained Joint Optimization: In (Chen et al., 19 Feb 2025), a complementarity constraint binds quadrotor and ground robot trajectories, ensuring the landing event is decided adaptively by both agents. This yields near real-time, globally feasible, collision-free rendezvous trajectories via a centralized, time-optimal MPC implemented with structure-exploiting solvers.
• Bi-Directional Attitude–Trajectory Coupling: The framework in (Zhao et al., 6 Jan 2026) positions the mobile platform as an active agent, capable of dynamically tilting its deck to a commanded optimal attitude ($\phi_\text{opt}$) derived from the UAV's optimal control problem. This parallelizes alignment and descent, breaking the sequential bottleneck of prior "track–then–descend" schemes, and is solved as a coupled nonlinear program that minimizes landing time subject to dynamic and actuation constraints.
• Minimal-Compute Control for Microscale Platforms: AIRA (Liu et al., 2024) relies on vector summation of photodiode outputs to obtain a light-gradient direction, feeding a proportional velocity controller. No explicit localization is performed; the system closes the guidance loop on this bearing estimate only.

Control Method	Key Principle	Notable Source
Bearing-State FSM	Minimal data, strong feedback	(Springer et al., 2024)
Complementarity MPC	Joint trajectory/event decision	(Chen et al., 19 Feb 2025)
Bi-directional Attitude	Parallelized landing process	(Zhao et al., 6 Jan 2026)
Photodiode P-control	Light-gradient vector summation	(Liu et al., 2024)

3. Fiducial Design, Detection, and Localization

Precision landing mandates robust and unambiguous pad identification under variable altitude, lighting, and occlusion.

• Visual Fiducials: AprilTag-based patterns remain prevalent, with concentric or multi-scale arrangements enabling detection at various approach altitudes (Springer et al., 2024, Salagame et al., 2022). Particular variants, such as AprilTag 36h11, are employed for their high detection reliability.
• Infrared Markers: To extend operation to night or visually-denied conditions, IR pads (active heated steel, or passive high-reflectivity aluminum) are used. These are paired with IR cameras or filtered photodiodes; image inversion may be performed to map IR intensity to tag bit-codes (Springer et al., 2024).
• IR Beacon-Based Localization: AIRA requires only a simple IR bulb on the pad and three drone-mounted photodiodes; the landing bearing is inferred solely by assignment of IR intensity gradients, without imaging or high-level detection (Liu et al., 2024).
• 3D Pose from PnP: Vision-based systems recover the tag pose via PnP (Perspective-n-Point) using known tag geometry and calibrated camera intrinsics. This enables full 6-DoF estimation when required (Salagame et al., 2022); however, some systems deliberately choose lower-dimensional, more robust bearing-only strategies to enhance noise immunity (Springer et al., 2024).

4. Precision, Robustness, and Experimental Results

Empirical studies across sampling platforms, environmental conditions, and mission types demonstrate centimeter-to-decimeter landing accuracies, with emphasis on robustness to occlusion, dynamic platform motion, and minimal onboard resources.

• Multimodal Fiducial System (Springer et al., 2024):
  • 26 autonomous landings in –8 °C to +3 °C, wind < 5 m/s.
  • Mean landing error: μ_E=0.19 m (σ_E=0.14 m); visual = 0.16 m, active IR = 0.14 m, passive IR = 0.26 m.
  • Successful recovery after induced occlusion; landing achieved after reacquisition.
• Joint Trajectory Planning (Chen et al., 19 Feb 2025):
  • Real-time optimization (<0.2 s) for N≈20–30 stages; hardware trials at 1.8 m initial separation, landing precision below 0.02 m, timing accuracy (landing event) constrained by progression variable.
• Bi-Directional Active Platforms (Zhao et al., 6 Jan 2026):
  • Sub-0.1 m landing error, terminal pitch error under 1°.
  • Platform speeds up to 2 m/s, landing times 2.4–4.6 s; order-of-magnitude speedup over prior methods for dynamic targets.
  • Empirical robustness to actuation lag and mission disturbances via real-time replanning.
• Minimal-Payload Photodiode System (AIRA, (Liu et al., 2024)):
  • Achieves mean error 9.2 cm (σ=4.5 cm) over range up to 11.1 m.
  • Outperforms monocular AprilTag/vision in both LOS and partial NLOS (92% NLOS trial success).
  • Sensing power an order of magnitude lower than camera-based approaches.

5. Challenges, Limitations, and Future Directions

Current cooperative landing systems confront several open challenges:

• Environmental Sensitivity: Weather and illumination variability (e.g., sky reflections affecting IR passive pads) can degrade tag detection reliability (Springer et al., 2024). Wind and platform motion induce control oscillations or marker loss (Salagame et al., 2022).
• Resource Constraints: For palm-sized platforms, power and payload severely limit sensor suite options. AIRA's approach is constrained by IR signal occlusion and single-emitter blind spots, suggesting that multi-emitter architectures or hybrid sensing will be necessary for robust, all-condition operation (Liu et al., 2024).
• Passive vs. Active Cooperation: Most previous systems considered the ground platform passive. Recent results demonstrate that active surface attitude manipulation on the ground agent substantially expands feasible UAV trajectories and enables parallel alignment and descent (Zhao et al., 6 Jan 2026).
• Trajectory Complexity: Heterogeneous multi-robot scenarios (e.g., mobile ground charging stations) necessitate coupled trajectory planning and tight time/space coordination, addressed via advanced complementarity constraint-based MPC formulations (Chen et al., 19 Feb 2025).
• Model Simplifications: Many cooperative schemes, including the bi-directional active platform approach, currently consider limited DoF in platform actuation (e.g., single-DOF tilt only). Extending to full 6-DoF deck motion would enhance generality for real-world vehicles (e.g., boats, uneven terrain) (Zhao et al., 6 Jan 2026).

A plausible implication is that integrating environmental observers (for wind, illumination), battery-aware optimization, and multi-agent task allocation will be required for resilient, large-scale deployment. Real-time sensor fusion (e.g., combining IMU, vision, IR, and platform telemetry) remains a focus.

6. Comparative Summary and Representative Results

System (arXiv ID/Year)	Sensing/Platform	Control/Planning	Landing Error	Special Features
(Springer et al., 2024) (2024)	Visual/IR, gimbal	FSM, bearing-only servo	0.19 m avg	Search–track–recover loop
(Chen et al., 19 Feb 2025) (2025)	OptiTrack, multi-robot	Complementarity MPC	<0.02 m	Joint trajectory/event
(Zhao et al., 6 Jan 2026) (2026)	Vision, platform deck	Coupled bi-directional opt	<0.1 m	Platform tilting, replanning
(Liu et al., 2024) (2024)	IR photodiode (micro)	Vector P-control	0.09 m	$<$0.1 mW, NLOS, indoor
(Salagame et al., 2022) (2022)	Monocular vision, tags	PID, staged tracking	0.15 m	Outdoor, moving target

Empirical results clearly demonstrate that coordinated, agent-coupled control protocols offer superior robustness, accuracy, and mission time scales over sequential or passive schemes, advancing applicability in dynamic, cost- and power-constrained environments.

7. Conclusion

Air–ground cooperative precision landing constitutes a rapidly advancing domain, leveraging diverse sensor modalities, minimal-data servoing, joint optimization, and bi-directional role allocation to realize robust, high-accuracy autonomous recovery. Experimental evidence from vision-based, IR-based, and multi-robot systems confirms sub-decimeter landing performance across static and mobile ground-agent configurations, with future progress expected in real-time planning, true multi-agent interaction, and operation under severe environmental and resource constraints (Springer et al., 2024, Chen et al., 19 Feb 2025, Zhao et al., 6 Jan 2026, Liu et al., 2024, Salagame et al., 2022).