Pith. sign in

REVIEW 3 major objections 5 minor 52 references

Hybrid Artificial Intelligence Strategies for Drone Navigation

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A drone navigation policy trained by reinforcement learning alone fails in obstacle courses, but pairing it with an expert rule-based engine raises task completion to 70% and eliminates collisions in the tested scenarios.

desk verdict Solid engineering result: the hybrid PPO-plus-rules architecture works in the tested simulated scenarios, but the obstacle-avoidance 'guarantee' is not actually a guarantee—the reported collision numbers are observed, not proven. read the letter →

arxiv 2501.04472 v1 pith:ELRV4RBK submitted 2025-01-08 cs.AI cs.RO

Rubén San-Segundo , Lucía Angulo , Manuel Gil-Martín , David Carramiñana , Ana M. Bernardos This is my paper · ORCID
classification cs.AIcs.RO
keywords dronenavigationreinforcementlearninghybridartificialintelligencerule-basedengineobstacleavoidanceexplainabilityhuman-in-the-loopPPO
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that a drone navigation policy learned by reinforcement learning is not enough when obstacles stand between drone and target, and that combining it with an expert rule-based engine yields a better system. In a simulated two-agent environment, the reinforcement-learning baseline completed only 15% of episodes and struck an obstacle in 100% of episodes; adding the rule-based obstacle-avoidance module raised completion to 70% and eliminated obstacle collisions entirely. For the target-searching task, the paper claims the hybrid saves about 20% of the time needed to find all targets compared with exhaustive search alone. The paper's overall conclusion is that hybrid AI — machine learning plus expert rules — gives a good compromise between performance and robustness, and that the same learned policy adapts to moving targets without retraining.

What carries the argument

The mechanism that carries the argument is the state-based switch between two policies. The system tracks an agent state: normal navigation (deep learning policy, PPO-Clip with a CNN feature extractor), stuck/obstacle state (rule-based engine), and exhaustive vs local search states. The specific rule for obstacle avoidance computes the obstacle center and the drone position, draws a circumference of radius equal to the drone-obstacle distance, and generates fictitious targets 30 degrees to the left; the drone steers toward these fictitious targets until a free path to the real target appears. This rule engine is what converts the failing reinforcement-learning policy into a successful one.

What would settle it

Run the hybrid system on a course where an obstacle ring encloses a target, or where two obstacles form a narrow corridor that the 30-degree fictitious-target rule cannot follow; if the drone fails to reach the target or collides in any such layout, the claim that the rule-based module guarantees the agents do not hit any obstacle does not hold.

Watch

Extended reading notes

Core claim

The central claim is that the navigation policy should not be left entirely to a deep reinforcement learning model. In the reaching-target task, the paper shows that a PPO-trained policy with reward shaping and a Manhattan-like distance measure can reach 100% of targets in obstacle-free settings, but with obstacles its task completion drops to 15%. Replacing the policy's behavior in 'stuck' states with a rule-based engine that generates fictitious targets on a circumference around the obstacle raises task success to 70% and reduces episodes with at least one obstacle collision from 100% to 0%. In the searching task, starting with a rule-based exhaustive search and switching to the reinforcement-learning model near a found target reduces the total cycles needed by roughly 20% across one to four target groups. The paper therefore claims hybrid AI is necessary and sufficient for robust performance in the scenarios tested.

Load-bearing premise

The whole collision-avoidance gain rests on the assumption that when a drone repeats movements without getting closer to the target, circling the obstacle along a circumference at a radius equal to the current distance, with fictitious targets 30 degrees apart, will always find an obstacle-free path; the paper itself notes that agents can still get stuck between several obstacles.

Editorial extensions

If this is right

  • Task completion with obstacles rises from 15% to 70% when the rule-based module handles stuck states, and obstacle collisions drop to 0% in the tested grid environment.
  • Reward shaping with distance-to-target and negative obstacle rewards reduces training cycles from 18 million to 6 million in the obstacle-free reaching task.
  • For searching, combining exhaustive rule-based search with RL local search cuts total cycles about 20% relative to exhaustive-only search for one to four target groups.
  • Adding a third dimension improves reaching-task success to around 92% with four obstacles because the drone has more escape paths.
  • The hybrid scheme requires no retraining when targets move; the same learned policy adapts, with task success around 90% in moving-target search.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the stuck-detection heuristic (repetitive movement without decreasing distance) is robust to noisy or partial observations in real flight, the same hybrid architecture could transfer to onboard drone controllers, where safety guarantees from rules complement learned agility.
  • A natural extension is to learn the recovery policy (when to switch and how to circle) rather than hand-coding fictitious targets, which could remove the paper's acknowledged remaining stuck-between-obstacles cases.
  • The 20% search-time saving depends on targets clustering in groups; in uniformly random target distributions the RL local-search advantage would likely shrink, so the reported gain is specific to the grouped-target assumption.
  • The evaluation methodology, with scenario-specific quality and time metrics, could serve as a template for comparing hybrid vs pure-learning navigation systems, but it would need randomized obstacle layouts and repeated runs to test the rule engine's geometric assumptions.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents a hybrid AI system for drone navigation in a simulated multi-agent environment, combining PPO-based deep reinforcement learning with a rule-based engine for obstacle avoidance and exhaustive search. Two tasks are considered: reaching known targets and searching for unknown targets. The main empirical results are that adding expert rules improves task success from 15% to 70% and eliminates collisions in the tested episodes for the reaching task, while for the searching task the hybrid reduces the number of cycles by about 20% in obstacle-free scenarios. The paper also describes explainability mechanisms (LIME, SHAP) and human interaction strategies. The overall conclusion is that hybrid AI combining machine learning and rule-based engines offers a good compromise between performance and robustness.

Significance. If the results are reproducible, this is a useful systems contribution to hybrid RL- and rule-based navigation, with a structured evaluation methodology and clearly specified metrics. The central comparison in Table 2 is internally consistent and supports the value of expert rules for collision avoidance. The use of standard tools (PettingZoo, Stable-Baselines3), the explicit acknowledgement of simplifications, and the description of the environment aid replication. However, the strength of the claims is limited by the lack of statistical validation and by a gap between the described geometric rule and its implementation, as detailed below.

major comments (3)
  1. [3.2.1, Figures 7-8] The obstacle-avoidance rule steps in straight lines toward fictitious targets placed 30 degrees apart on a circumference; each straight step is a chord that can approach the obstacle center at distance R*cos(15°) ≈ 0.966R. If the drone's initial clearance is less than about 3.4% of R, the chord enters the obstacle's forbidden zone, so the rule does not guarantee a collision-free trajectory. The stuck-detection condition ('repetitive movements in a cycle without decreasing distance') does not enforce any minimum clearance. Therefore the statement in Section 4.2.2 that expert rules are 'the only way to guarantee that the agents do not hit any obstacle' is too strong, and the 0%-collision row in Table 2 is an observed outcome on the tested episodes, not a guarantee of the stated algorithm. I recommend either modifying the rule (e.g., following the arc instead of chords, or requiring a clearance margin larger than the chord sagitta) or rephrasing the claim as an empirical result on the evaluated scenarios.
  2. [Section 4, Tables 1-5] The evaluation reports point estimates over 200 episodes with no information about seeds, variance, confidence intervals, or statistical significance. The conclusions that the hybrid yields a 'significant reduction' of about 20% in cycles and is a 'very good compromise' are therefore not statistically supported; with a single run, the observed differences could be due to randomness. Please provide results over multiple seeds with confidence intervals or a significance test, and temper the wording of the conclusions accordingly.
  3. [Section 4.3.2, Table 5 and Abstract] The claimed 20% reduction in time is not representative of all scenarios. With obstacles and four target groups, the hybrid system uses more cycles (1299.3) than the baseline exhaustive search (1243.3), and for one and three groups the reductions are only about 4.3% and 4.9%. The abstract's conclusion that 'it has been possible to reduce 20% of the time required to locate all the targets' applies only to obstacle-free settings and should be conditioned on scenario, or the abstract should report the range of reductions observed.
minor comments (5)
  1. [3.2.1] The word 'radio' should be 'radius' in the text describing the circumference around the obstacle.
  2. [3.3] The word 'fictious' appears in Section 3.2.1 and Section 3.3; it should be 'fictitious'.
  3. [Tables 3-5] The row labels for the number of target groups are implicit; please state explicitly which row corresponds to one, two, three, and four groups.
  4. [3.1] The observation space is described as a '20 20 submatrix'; please use '20 × 20' for clarity.
  5. [4.2.3] Figure 17 is referenced but the figure itself is not visible in the text; please ensure the figure is legible and the axes are labeled so the reader can verify the described percentages.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper's claims are empirical evaluations of hand-designed hybrid policies, not derivations from their own inputs.

full rationale

The paper makes no equation-level derivation from which its central results follow by construction. The main claims are empirical measurements: Table 2 reports that adding expert rules raises task completion from 15% to 70% and reduces collision episodes from 100% to 0% over 200 newly generated episodes, and Section 5 generalizes from these observed results. The rule-based obstacle-avoidance mechanism in Section 3.2.1 is a hand-designed heuristic (fictitious targets around a computed circumference), and its performance is evaluated rather than assumed, so there is no fitted-parameter-renamed-as-prediction pattern. Reward shaping, PPO hyperparameters, and the fictitious-target angle are disclosed design choices, not parameters fitted to the reported test metrics. The paper contains no load-bearing self-citations: cited references are external RL toolkits, PPO, LIME, SHAP, and background path-planning works, and no uniqueness theorem is imported from the authors' prior work. The paper even acknowledges failure cases, e.g., agents getting stuck between several obstacles and targets placed in obstacle security zones, which confirms the results are contingent empirical observations rather than forced identities. A geometric concern about chord steps vs. arcs in the obstacle-avoidance rule is a correctness or generalization risk, not circularity. Overall, the derivation chain is self-contained and empirically grounded, so the circularity score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 1 invented entities

The central results rest on hand-tuned reward weights and PPO hyperparameters, an unverified simulator abstraction, and the geometric obstacle-avoidance heuristic. None of these are independently validated, so the empirical gains should be read as specific to this simulator and configuration.

free parameters (5)
  • PPO learning rate = 0.0003
    Adjusted in preliminary experiments (Sections 4.2.1-4.3.2); the RL comparisons depend on this choice.
  • PPO clip range = 0.3
    Set following Stable-Baselines3 recommendation and preliminary experiments (Section 3.2.1).
  • Reward shaping weights = T - O - DeltaD with unit weights
    Hand-chosen reward coefficients define the learned policy objective; no ablation on the coefficients is reported (Sections 3.1 and 4.2.2).
  • Fictitious target angular step = 30 degrees
    Rule-based obstacle avoidance places waypoints 30 degrees apart on a circumference around the obstacle (Section 3.2.1); no sensitivity analysis is provided.
  • Stuck detection threshold = not specified ('several movements')
    The switch to rule-based mode triggers after repeated non-progressing movements; the exact count is not reported (Section 3.2.1).
assumptions (4)
  • domain assumption The PettingZoo and Stable-Baselines3 library implementations used are correct enough for the reported conclusions.
    The system relies on these libraries (Sections 3.1, 3.2.1) without independent verification of their internals.
  • domain assumption A drone detects a target when both occupy the same grid cell.
    Explicitly stated in the Conclusions as a simplification; the task success rates depend on this detection rule.
  • domain assumption The circumference-based fictitious-target loop always produces a collision-free route around an obstacle.
    Section 3.2.1 assumes this geometric strategy works; Table 2 shows residual failure cases, so the assumption is not fully satisfied.
  • standard math Distance-based reward shaping does not change the optimal policy, only the learning speed.
    The paper uses distance-reduction partial rewards (Section 3.1) without proving policy invariance; this is a standard RL assumption.
invented entities (1)
  • Fictitious target
    purpose: A waypoint generated by the rule-based engine to steer the drone around an obstacle.
    This is an algorithmic construct, not a physical entity; its behavior is fully defined by the authors and no external evidence is claimed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hybrid Artificial Intelligence Strategies for Drone Navigation." pith.science (2026). https://pith.science/paper/ELRV4RBK

@misc{pith2026250104472,
  author       = {Pith},
  title        = {Pith review of: Hybrid Artificial Intelligence Strategies for Drone Navigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ELRV4RBK}},
  note         = {Machine review of arXiv:2501.04472}
}
read the original abstract

Objective: This paper describes the development of hybrid artificial intelligence strategies for drone navigation. Methods: The navigation module combines a deep learning model with a rule-based engine depending on the agent state. The deep learning model has been trained using reinforcement learning. The rule-based engine uses expert knowledge to deal with specific situations. The navigation module incorporates several strategies to explain the drone decision based on its observation space, and different mechanisms for including human decisions in the navigation process. Finally, this paper proposes an evaluation methodology based on defining several scenarios and analyzing the performance of the different strategies according to metrics adapted to each scenario. Results: Two main navigation problems have been studied. For the first scenario (reaching known targets), it has been possible to obtain a 90% task completion rate, reducing significantly the number of collisions thanks to the rule-based engine. For the second scenario, it has been possible to reduce 20% of the time required to locate all the targets using the reinforcement learning model. Conclusions: Reinforcement learning is a very good strategy to learn policies for drone navigation, but in critical situations, it is necessary to complement it with a rule-based module to increase task success rate.

Figures

Figures reproduced from arXiv: 2501.04472 by the authors.

Figure 1
Figure 1. Diagram of the global system. The global system is composed of two main modules: The agent (drone) module generates the next drone action using the local observation of the agent. This process selects the drone’s action proposed from the rule-based module or the deep learning model, depending on the agent state. The input of this module is the local observation space (a 20 20 submatrix, explained in Section 3.1). Th… view at source ↗
Figure 2
Figure 2. Visual representation of the multiagent environment. In this work, two navigation tasks (reaching located targets and searching for new targets) have been considered with several scenarios per task. The objective of analyzing several scenarios per task is to have a more complete analysis covering a wider range of applications. 2.1. Reaching Located Targets In this task, the drones must navigate through an environmen… view at source ↗
Figure 3
Figure 3. Environment state representation: a 200 × 200 matrix of the whole environment. Reward strategy The reward strategy includes the following two main aspects: Targetreached by an agent or drone: when a target isreached by an agent, the corre￾sponding agent’sreward isincreased by 1, emphasizing the importance of reaching a target. Partial rewards depending on the distance to the target: An additional reward based on the… view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Alternative paths with the same distance. 3.2. Agent Module The main target of this module is to define the next action according to the drone navigation policy. The next drone action is selected from the rule-based module or from the deep learning model depending on t…
Figure 5
Figure 5. Figure 5: Observation space representation: a 20 × 20 submatrix. • [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Deep learning architecture for the policy model [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Path followed by the drone to avoid the obstacle thanks to the rule-based module. The process starts when an obstacle is detected between the agent and the target, and the drone is not able to find an alternative path without obstacles ( [PITH_FULL_IMAGE:figures/full_…
Figure 8
Figure 8. Figure 8: Flowchart for avoiding an obstacle in the path to the target [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Flowchart for exhaustive search: drone movement decision. Priority of movements [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Exhaustive search in vertical paths (arrows indicate drone’s direction) [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 11
Figure 11. Figure 11: shows the local observation, a matrix where the agent’s observation space is represented. This figure shows two obstacles corresponding to the black blocks in the next image (bottom and right parts) [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]
Figure 12
Figure 12. Figure 12: Analyses of actions 1 (forward), 2 (backward), 3 (left), 4 (right) in a 2D environment. 3.4.2. SHapley Additive exPlanations(SHAP) Additionally, the study also includes an analysis using SHAP to obtain more details about the explainability of the decisions made by the…
Figure 14
Figure 14. Figure 14: Visualization of manual movement of agents and manual change of agent’s target (right button of the mouse) [PITH_FULL_IMAGE:figures/full_fig_p014_14.png]
Figure 15
Figure 15. Figure 15: Manual change of agent’s target (right button of the mouse) [PITH_FULL_IMAGE:figures/full_fig_p014_15.png]
Figure 16
Figure 16. Figure 16: Visualization of the information window and drone positions associated with the visual￾ization window. Human interaction strategies have been used for recreating specific situations that can be debugged using the explainability mechanisms described above. Human intera…
Figure 17
Figure 17. Figure 17: Main results when reaching targets with obstacles, considering a 3D scenario. Total used cycles in training is 8 · 106, and the maximum number of cycles per episode in testing is 200. 4.3. Task 2: Searching for Targets This section presents the results obtained after …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 47 canonical work pages

  1. [1]

    Bernardos 2 1 Speech Technology and Machine Learning Group, Information Processing and Telecommunications Center, Universidad Politécnica de Madrid, ETSI Telecomunicación, Av

    Article Hybrid Artificial Intelligence Strategies for Drone Navigation Rubén San-Segundo 1,* , Lucía Angulo 1, Manuel Gil-Martín 1,* , David Carramiñana 2 and Ana M. Bernardos 2 1 Speech Technology and Machine Learning Group, Information Processing and Telecommunications Center, Universidad Politécnica de Madrid, ETSI Telecomunicación, Av. Complutense, 30...

    2024

  2. [2]

    This way, it is possible to simulate 2D and 3D navigation scenarios

    Every object includes a Z coordinate in black, indicating its height. This way, it is possible to simulate 2D and 3D navigation scenarios. The inputs to this module are the drones’ actions, and the outputs are the new environment state and the reward associated with every action. This reward is used only during training. Additionally, the system integrate...

    work page 2024

  3. [3]

    Environment state representation: a 200 × 200 matrix of the whole environment. Reward strategy The reward strategy includes the following two main aspects: Target reached by an agent or drone: when a target is reached by an agent, the corre- sponding agent’s reward is increased by 1, emphasizing the importance of reaching a target. Partial rewards dependi...

    work page 2024

  4. [4]

    In this work, two navigation tasks (reaching located targets and searching for new targets) have been considered with several scenarios per task

    Visual representation of the multiagent environment. In this work, two navigation tasks (reaching located targets and searching for new targets) have been considered with several scenarios per task. The objective of analyzing several scenarios per task is to have a more complete analysis covering a wider range of applications. 2.1. Reaching Located Target...

    work page 2024

  5. [5]

    • AI 2024, 5 2110 The next sections describe the different hybrid strategies depending on the navigation task

    Observation space representation: a 20 × 20 submatrix. • AI 2024, 5 2110 The next sections describe the different hybrid strategies depending on the navigation task. 3.2.1. Task 1: Reaching Located Targets In this task, the drones must navigate to the different targets until reaching all of them. Every drone is assigned a specific target that it must reac...

    work page 2024

  6. [6]

    and a significant impact across a great variability of sectors ranging from healthcare [7], smart cities [8], natural language processing and human–computer interac- tion [9], to transportation and logistics [10]. Similarly, these algorithms have an important role in drone applications because ML enhances drones’ capabilities in navigation [11], object de...

    work page 2024

  7. [7]

    This behavior is provoked by changing the agent target; the system generates a sequence of fictitious targets that go around the obstacle. • • AI 2024, 5 2111 While (the drone cannot find a path: repetitive movements in a cycle) Estimation of the center of the obstacle (cx, cy, cz) and drone position (dx, dy, dz) Computation of a circumference around the ...

    work page 2024

  8. [8]

    AI 2024, 5 2112 As shown, the algorithm is continuously (at each step) evaluating the existence of an obstacle between the drone and the target

    Flowchart for avoiding an obstacle in the path to the target. AI 2024, 5 2112 As shown, the algorithm is continuously (at each step) evaluating the existence of an obstacle between the drone and the target. If the target or the obstacle changed their position, the algorithm was able to adapt itself to this new situation. 3.2.2. Task 2: Searching for Targe...

    work page 2024

Show all 52 references
  1. [9]

    Deep Transfer Learning for Automatic Speech Recognition: Towards Better Generalization

    Kheddar, H.; Himeur, Y.; Al-Maadeed, S.; Amira, A.; Bensaali, F. Deep Transfer Learning for Automatic Speech Recognition: Towards Better Generalization. arXiv 2023, arXiv:2304.14535. [CrossRef]

    2023 arXiv

  2. [10]

    Exhaustive search in vertical paths (arrows indicate drone’s direction). AI 2024, 5 2113 × − − Second state: local search around the last target detected When one target is detected, the drone stops the exhaustive search, and a reinforcement learning model guides the drone mov...

    2024

  3. [11]

    • • AI 2024, 5 2115 Figure

    2024

  4. [12]

    Analyses of actions 1 (forward), 2 (backward), 3 (left), 4 (right) in a 2D environment. 3.4.2. SHapley Additive exPlanations (SHAP) Additionally, the study also includes an analysis using SHAP to obtain more details about the explainability of the decisions made by the agents/...

    2024

  5. [13]

    A hybrid optimization framework for UAV reconnaissance mission planning

    Liu, W.; Zhang, T.; Huang, S.; Li, K. A hybrid optimization framework for UAV reconnaissance mission planning. Comput. Ind. Eng. 2022, 173, 108653. [CrossRef]

    2022

  6. [14]

    Information window

    and the target (right side) of the agents (Figure 15). Information window. The following image shows the visualization of the information window that shows the coordinates of each agent. These coordinates can be modified manually. Figure 16, at the bottom, shows the observatio...

    2024

  7. [15]

    It is important to remark that human interaction strategies (described in the previous section) have not been used during the system evaluation and testing

    Evaluation Results and Discussions This section describes the general evaluation methodology, and the analyses carried out for both tasks: reaching several located targets and searching for targets. It is important to remark that human interaction strategies (described in the ...

    2024

  8. [16]

    Human interaction strategies have been used for recreating specific situations that can be debugged using the explainability mechanisms described above

    Visualization of the information window and drone positions associated with the visual- ization window. Human interaction strategies have been used for recreating specific situations that can be debugged using the explainability mechanisms described above. Human interaction st...

    2024

  9. [17]

    Main results when reaching known targets with obstacles, considering a 2D scenario. Maximum Number of Cycles Per Episode in Testing % of Cycles Using Deep Learning (DL) or Rule-Based (RB) % of Episodes Reaching All Targets (Task Success) % of Episodes with (at Least) One Drone...

    2024

  10. [18]

    Different number of groups

    Results regarding the different implementations for searching random static target without obstacles. Different number of groups. Total Cycles in Training Cycles per Episode in Testing: Initial Exhaustive Search Cycles per Episode in Testing: RL Search Cycles per Episode in Te...

    2024

  11. [19]

    Results include movements of the targets trying to reach the bottom of the scene to escape from the scene. Total Cycles Used in Training Cycles per Episode in Testing: Initial Exhaustive Search Cycles per Episode in Testing: RL Search Cycles per Episode in Testing: Posterior E...

    2024

  12. [20]

    The hybrid AI combines deep learning models with rule-based strategies to generate the agent action based on the agent state

    Conclusions This paper describes the development and evaluation of hybrid artificial intelligence strategies for drone navigation in simulated environments. The hybrid AI combines deep learning models with rule-based strategies to generate the agent action based on the agent s...

  13. [21]

    UAV-assisted data collection for internet of things: A survey

    Wei, Z.; Zhu, M.; Zhang, N.; Wang, L.; Zou, Y.; Meng, Z.; Wu, H.; Feng, Z. UAV-assisted data collection for internet of things: A survey. IEEE Internet Things J. 2022, 9, 15460–15483. [CrossRef]

    2022

  14. [22]

    Vision Based Drone Obstacle Avoidance by Deep Reinforcement Learning

    Xue, Z.; Gonsalves, T. Vision Based Drone Obstacle Avoidance by Deep Reinforcement Learning. AI 2021, 2, 366–380. [CrossRef]

    2021

  15. [23]

    Routing protocols for Unmanned Aerial Vehicle Networks: A survey

    Arafat, M.Y.; Moh, S. Routing protocols for Unmanned Aerial Vehicle Networks: A survey. IEEE Access 2019, 7, 99694–99720. [CrossRef]

    2019

  16. [24]

    Topology control algorithms in multi-unmanned aerial vehicle networks: An extensive survey

    Alam, M.M.; Arafat, M.Y.; Moh, S.; Shen, J. Topology control algorithms in multi-unmanned aerial vehicle networks: An extensive survey. J. Netw. Comput. Appl. 2022, 207, 103495. [CrossRef]

    2022

  17. [25]

    Vision-Based Navigation Techniques for Unmanned Aerial Vehicles: Review and Challenges

    Arafat, M.Y.; Alam, M.M.; Moh, S. Vision-Based Navigation Techniques for Unmanned Aerial Vehicles: Review and Challenges. Drones 2023, 7,

    2023

  18. [26]

    AI Advancements: Comparison of Innovative Techniques

    Taherdoost, H.; Madanchian, M. AI Advancements: Comparison of Innovative Techniques. AI 2024, 5, 38–54. [CrossRef]

    2024

  19. [27]

    Face mask detection in smart cities using deep and transfer learning: Lessons learned from the COVID-19 pandemic

    Himeur, Y.; Al-Maadeed, S.; Varlamis, I.; Al-Maadeed, N.; Abualsaud, K.; Mohamed, A. Face mask detection in smart cities using deep and transfer learning: Lessons learned from the COVID-19 pandemic. Systems 2023, 11,

    2023

  20. [28]

    Available online: https://pettingzoo.farama.org/index.html (accessed on 7 September 2024)

    PettingZoo Library. Available online: https://pettingzoo.farama.org/index.html (accessed on 7 September 2024)

    2024

  21. [29]

    Performance and energy optimization of building automation and management systems: Towards smart sustainable carbon-neutral sports facilities

    Elnour, M.; Fadli, F.; Himeur, Y.; Petri, I.; Rezgui, Y.; Meskin, N.; Ahmad, A.M. Performance and energy optimization of building automation and management systems: Towards smart sustainable carbon-neutral sports facilities. Renew. Sustain. Energy Rev. 2022, 162, 112401. [CrossRef]

    2022

  22. [30]

    Proximal Policy Optimization Algorithms

    Schulman, J.; Wolski, F.; Dhariwal, P.; Radford, A.; Klimov, O. Proximal Policy Optimization Algorithms. arXiv 2017, arXiv:1707.06347

    2017 arXiv

  23. [31]

    Smart Speed Camera Based on Automatic Number Plate Recognition for Residential Compounds and Institutions Inside Qatar

    Al-Hasan, T.M.; Shibeika, A.S.; Attique, U.; Bensaali, F.; Himeur, Y. Smart Speed Camera Based on Automatic Number Plate Recognition for Residential Compounds and Institutions Inside Qatar. In Proceedings of the 2022 5th International Conference on Signal Processing and Inform...

    2022

  24. [32]

    On the achievability of submeter-accurate UAV navigation with cellular signals exploiting loose network synchronization

    Khalife, J.; Kassas, Z.M. On the achievability of submeter-accurate UAV navigation with cellular signals exploiting loose network synchronization. IEEE Trans. Aerosp. Electron. Syst. 2022, 58, 4261–4278. [CrossRef]

    2022

  25. [33]

    Panoptic segmentation: A review

    Elharrouss, O.; Al-Maadeed, S.; Subramanian, N.; Ottakath, N.; Almaadeed, N.; Himeur, Y. Panoptic segmentation: A review. arXiv 2021, arXiv:2111.10250

    2021 arXiv

  26. [35]

    A review of GNSS-independent UAV navigation techniques

    Gyagenda, N.; Hatilima, J.V.; Roth, H.; Zhmud, V. A review of GNSS-independent UAV navigation techniques. Robot. Auton. Syst. 2022, 152, 104069. [CrossRef]

    2022

  27. [36]

    A Formal Basis for the Heuristic Determination of Minimum Cost Paths

    Hart, P.; Nilsson, N.; Raphael, B. A Formal Basis for the Heuristic Determination of Minimum Cost Paths. IEEE Trans. Syst. Sci. Cybern. 1968, 4, 100–107. [CrossRef]

    1968

  28. [37]

    Rapidly-exploring random trees: Progress and prospects

    LaValle, S.M.; Kuffner, J.J.; Donald, B. Rapidly-exploring random trees: Progress and prospects. Algorithm. Comput. Robot. New Dir. 2001, 5, 293–308

    2001

  29. [38]

    Navigation and Deployment of Solar-Powered Unmanned Aerial Vehicles for Civilian Applications: A Comprehensive Review

    Li, S.; Fang, Z.; Verma, S.C.; Wei, J.; Savkin, A.V. Navigation and Deployment of Solar-Powered Unmanned Aerial Vehicles for Civilian Applications: A Comprehensive Review. Drones 2024, 8,

    2024

  30. [39]

    Reinforcement learning: A survey

    Kaelbling, L.P.; Littman, M.L.; Moore, A.W. Reinforcement learning: A survey. J. Artif. Intell. Res. 1996, 4, 237–285. [CrossRef]

    1996

  31. [40]

    Ostrovski Human-level control through deep reinforcement learning

    Mnih, V.; Kavukcuoglu, K.; Silver, D.; Rusu, A.A.; Veness, J.; Bellemare, M.G.; Graves, A.; Riedmiller, M.; Fidjeland, A.K. Ostrovski Human-level control through deep reinforcement learning. Nature 2015, 518, 529–533. [CrossRef]

    2015

  32. [41]

    Optimization Strategies for Atari Game Environments: Integrating Snake Optimization Algorithm and Energy Valley Optimization in Reinforcement Learning Models

    Sarkhi, S.M.K.; Koyuncu, H. Optimization Strategies for Atari Game Environments: Integrating Snake Optimization Algorithm and Energy Valley Optimization in Reinforcement Learning Models. AI 2024, 5, 1172–1191. [CrossRef]

    2024

  33. [42]

    Autonomous UAV navigation using reinforcement learning

    Pham, H.X.; La, H.M.; Feil-Seifer, D.; Nguyen, L.V. Autonomous UAV navigation using reinforcement learning. arXiv 2018, arXiv:1801.05086

    2018 arXiv

  34. [43]

    Deep reinforcement learning for drone navigation using sensor data

    Hodge, V.J.; Hawkins, R.; Alexander, R. Deep reinforcement learning for drone navigation using sensor data. Neural Comput. Applic. 2021, 33, 2015–2033. [CrossRef]

    2021

  35. [44]

    Deep reinforcement learning for drone delivery

    Muñoz, G.; Barrado, C.; Çetin, E.; Salami, E. Deep reinforcement learning for drone delivery. Drones 2019, 3,

    2019

  36. [45]

    Learning to adapt in dynamic, real-world environments through meta-reinforcement learning

    Nagabandi, A.; Clavera, I.; Liu, S.; Fearing, R.S.; Abbeel, P.; Levine, S.; Finn, C. Learning to adapt in dynamic, real-world environments through meta-reinforcement learning. arXiv 2018, arXiv:1803.11347

    2018 arXiv

  37. [46]

    Beyond Static Obstacles: Integrating Kalman Filter with Reinforcement Learning for Drone Navigation

    Marino, F.; Guglieri, G. Beyond Static Obstacles: Integrating Kalman Filter with Reinforcement Learning for Drone Navigation. Aerospace 2024, 11,

    2024

  38. [47]

    Multi-objective reinforcement learning for autonomous drone navigation in urban areas with wind zones

    Wu, J.; Ye, Y.; Du, J. Multi-objective reinforcement learning for autonomous drone navigation in urban areas with wind zones. Autom. Constr. 2024, 158, 105253. [CrossRef]

    2024

  39. [49]

    Available online: https://stable-baselines3.readthedocs.io/en/master/ (accessed on 7 September 2024)

    Stable-Baselines3 RL Toolkit. Available online: https://stable-baselines3.readthedocs.io/en/master/ (accessed on 7 September 2024)

    2024

  40. [51]

    Available online: https://github.com/marcotcr/lime (accessed on 7 September 2024)

    LIME Python Library. Available online: https://github.com/marcotcr/lime (accessed on 7 September 2024)

    2024

  41. [52]

    Available online: https://github.com/shap/shap/blob/master/docs/index.rst (accessed on 7 September 2024)

    SHAP Python Library. Available online: https://github.com/shap/shap/blob/master/docs/index.rst (accessed on 7 September 2024). Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contribu...

    2024

  42. [106]

    Including a negative reward when hitting an obstacle is crucial to reduce the number of situations where one drone hits an obstacle (second row)

    The main conclusions from these experiments are the following: When including the obstacles, it is necessary to increase the number of cycles in training to better learn the agent policy (compared to Table 1). Including a negative reward when hitting an obstacle is crucial to ...

    2024

  43. [107]

    [CrossRef] AI 2024, 5 2126

    2024

  44. [200]

    4.3. Task 2: Searching for Targets This section presents the results obtained after evaluating the different strategies implemented for the second task: the agents (drones) must go over all the space searching for targets. This scenario includes several situations including or...

    2024

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.